Merge branch '3.0' of https://github.com/taosdata/TDengine into refact/tsdb_last

2022-08-09 11:59:40 +00:00 · 2022-08-09 11:59:40 +00:00 · d242226d1a
parent 4ac14697b7 b57e6c15ec
commit d242226d1a
100 changed files with 1810 additions and 607 deletions
--- a/README-CN.md
+++ b/README-CN.md
@ -50,19 +50,12 @@ TDengine 目前 2.0 版服务器仅能在 Linux 系统上安装和运行，后

 ## 安装工具

-### Ubuntu 16.04 及以上版本 & Debian：
+### Ubuntu 18.04 及以上版本 & Debian：

 ```bash
 sudo apt-get install -y gcc cmake build-essential git libssl-dev
 ```

-### Ubuntu 14.04：
-
-```bash
-sudo apt-get install -y gcc cmake3 build-essential git binutils-2.26
-export PATH=/usr/lib/binutils-2.26/bin:$PATH
-```
-
 编译或打包 JDBC 驱动源码，需安装 Java JDK 8 或以上版本和 Apache Maven 2.7 或以上版本。

 安装 OpenJDK 8：
@ -89,7 +82,7 @@ taosTools 是用于 TDengine 的辅助工具软件集合。目前它包含 taosB
 sudo apt install build-essential libjansson-dev libsnappy-dev liblzma-dev libz-dev pkg-config
 ```

-### CentOS 7：
+### CentOS 7.9：

 ```bash
 sudo yum install -y gcc gcc-c++ make cmake git openssl-devel
@ -161,12 +154,7 @@ git clone https://github.com/taosdata/TDengine.git
 cd TDengine
 ```

-Go 连接器和 Grafana 插件在其他独立仓库，如果安装它们的话，需要在 TDengine 目录下通过此命令安装：
-
-```bash
-git submodule update --init --recursive
-```
-
+Go 连接器和 Grafana 插件已移到其他独立仓库。
 如果使用 https 协议下载比较慢，可以通过修改 ~/.gitconfig 文件添加以下两行设置使用 ssh 协议下载。需要首先上传 ssh 密钥到 GitHub，详细方法请参考 GitHub 官方文档。

 ```
@ -187,7 +175,6 @@ git submodule update --init --recursive
 这个脚本等价于执行如下命令：

 ```bash
-git submodule update --init --recursive
 mkdir debug
 cd debug
 cmake .. -DBUILD_TOOLS=true
--- a/README.md
+++ b/README.md
@ -53,19 +53,12 @@ To build TDengine, use [CMake](https://cmake.org/) 3.0.2 or higher versions in t

 ## Install build dependencies

-### Ubuntu 16.04 and above or Debian
+### Ubuntu 18.04 and above or Debian

 ```bash
 sudo apt-get install -y gcc cmake build-essential git libssl-dev
 ```

-### Ubuntu 14.04
-
-```bash
-sudo apt-get install -y gcc cmake3 build-essential git binutils-2.26
-export PATH=/usr/lib/binutils-2.26/bin:$PATH
-```
-
 To compile and package the JDBC driver source code, you should have a Java jdk-8 or higher and Apache Maven 2.7 or higher installed.

 To install openjdk-8:
@ -91,7 +84,7 @@ To build the [taosTools](https://github.com/taosdata/taos-tools) on Ubuntu/Debia
 sudo apt install build-essential libjansson-dev libsnappy-dev liblzma-dev libz-dev pkg-config
 ```

-### CentOS 7
+### CentOS 7.9

 ```bash
 sudo yum install epel-release
@ -164,12 +157,7 @@ git clone https://github.com/taosdata/TDengine.git
 cd TDengine
 ```

-The connectors for go & Grafana and some tools have been moved to separated repositories,
-so you should run this command in the TDengine directory to install them:
-
-```bash
-git submodule update --init --recursive
-```
+The connectors for go & Grafana and some tools have been moved to separated repositories.

 You can modify the file ~/.gitconfig to use ssh protocol instead of https for better download speed. You need to upload ssh public key to GitHub first. Please refer to GitHub official documentation for detail.

@ -191,7 +179,6 @@ You can run the bash script `build.sh` to build both TDengine and taosTools incl
 It equals to execute following commands:

 ```bash
-git submodule update --init --recursive
 mkdir debug
 cd debug
 cmake .. -DBUILD_TOOLS=true
--- a/build.sh
+++ b/build.sh
@ -0,0 +1,8 @@
+#!/bin/bash
+
+if [ ! -d debug ]; then
+    mkdir debug || echo -e "failed to make directory for build"
+fi
+
+cd debug && cmake .. -DBUILD_TOOLS=true && make
+
--- a/cmake/cmake.define
+++ b/cmake/cmake.define
@ -115,9 +115,7 @@ ELSE ()
    ENDIF ()

    MESSAGE("System processor ID: ${CMAKE_SYSTEM_PROCESSOR}")
-    IF (${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64")
-        ADD_DEFINITIONS("-D_TD_ARM_")
-    ELSE ()
+    IF (TD_INTEL_64 OR TD_INTEL_32)
        ADD_DEFINITIONS("-msse4.2")
        IF("${FMA_SUPPORT}" MATCHES "true")
            MESSAGE(STATUS "turn fma function support on")
--- a/cmake/cmake.platform
+++ b/cmake/cmake.platform
@ -85,26 +85,36 @@ IF ("${CPUTYPE}" STREQUAL "")
    MESSAGE(STATUS "The current platform is aarch32")
    SET(PLATFORM_ARCH_STR "arm")
    SET(TD_ARM_32 TRUE)
+    ADD_DEFINITIONS("-D_TD_ARM_")
+    ADD_DEFINITIONS("-D_TD_ARM_32")
  ELSEIF (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
    MESSAGE(STATUS "The current platform is aarch64")
    SET(PLATFORM_ARCH_STR "arm64")
    SET(TD_ARM_64 TRUE)
+    ADD_DEFINITIONS("-D_TD_ARM_")
+    ADD_DEFINITIONS("-D_TD_ARM_64")
  ENDIF ()
 ELSE ()
  # if generate ARM version:
  #   cmake -DCPUTYPE=aarch32 .. or cmake -DCPUTYPE=aarch64
-  IF (${CPUTYPE} MATCHES "aarch32")
+  IF (${CPUTYPE} MATCHES "aarch32" or ${CPUTYPE} MATCHES "arm32")
    SET(PLATFORM_ARCH_STR "arm")
    MESSAGE(STATUS "input cpuType: aarch32")
+    ADD_DEFINITIONS("-D_TD_ARM_")
+    ADD_DEFINITIONS("-D_TD_ARM_32")
    SET(TD_ARM_32 TRUE)
-  ELSEIF (${CPUTYPE} MATCHES "aarch64")
+  ELSEIF (${CPUTYPE} MATCHES "aarch64" or ${CPUTYPE} MATCHES "arm64")
    SET(PLATFORM_ARCH_STR "arm64")
    MESSAGE(STATUS "input cpuType: aarch64")
+    ADD_DEFINITIONS("-D_TD_ARM_")
+    ADD_DEFINITIONS("-D_TD_ARM_64")
    SET(TD_ARM_64 TRUE)
  ELSEIF (${CPUTYPE} MATCHES "mips64")
    SET(PLATFORM_ARCH_STR "mips")
    MESSAGE(STATUS "input cpuType: mips64")
    SET(TD_MIPS_64 TRUE)
+    ADD_DEFINITIONS("-D_TD_MIPS_")
+    ADD_DEFINITIONS("-D_TD_MIPS_64")
  ELSEIF (${CPUTYPE} MATCHES "x64")
    SET(PLATFORM_ARCH_STR "amd64")
    MESSAGE(STATUS "input cpuType: x64")
--- a/cmake/taosws_CMakeLists.txt.in
+++ b/cmake/taosws_CMakeLists.txt.in
@ -1,8 +1,8 @@

 # taosws-rs
 ExternalProject_Add(taosws-rs
-        GIT_REPOSITORY https://github.com/taosdata/taosws-rs.git
-        GIT_TAG 29424d5
+        GIT_REPOSITORY https://github.com/taosdata/taos-connector-rust.git
+        GIT_TAG 97c4bac
        SOURCE_DIR "${TD_SOURCE_DIR}/tools/taosws-rs"
        BINARY_DIR ""
        #BUILD_IN_SOURCE TRUE
--- a/docs/assets/tdengine.svg
+++ b/docs/assets/tdengine.svg
@ -0,0 +1,44 @@
+<?xml version="1.0" encoding="utf-8"?>
+<svg version="1.1" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" x="0px" y="0px"
+	 viewBox="0 0 500 105" style="enable-background:new 0 0 1000 210;" xml:space="preserve">
+<style type="text/css">
+	.st0{fill:#18499D;}
+	.st1{fill-rule:evenodd;clip-rule:evenodd;fill:#18499D;}
+</style>
+<g>
+	<path class="st0" d="M153.2,27.1h-16v49h-11.1v-49h-15.9v-9.7h43L153.2,27.1z"/>
+	<path class="st0" d="M180.6,17.4c17.1,0,28.5,4.9,28.5,29.4c0,22.7-11.4,29.4-28.5,29.4h-19V17.4H180.6z M180.1,66.5
+		c11.3,0,17.6-4.3,17.6-19.8c0-16.3-6.5-20-17.6-20h-7.4v39.8L180.1,66.5z"/>
+	<path class="st0" d="M259.4,55.3l-30,2c0.2,7.7,4,11.2,13.1,11.2c5.1-0.1,10.2-1.1,14.9-3.1V74c-2.7,1.6-9.3,3-16.8,3
+		c-13.9,0-21.9-5.2-21.9-23.7s8-23.7,21.9-23.7c15.8,0,19.1,9,19.1,19.7C259.8,51.3,259.7,53.3,259.4,55.3z M250,48.2
+		c0-5.7-1.2-10.5-9.3-10.5c-8.5,0-11,4.1-11.3,12L250,48.2z"/>
+	<path class="st0" d="M310.8,44.8v31.3h-10.6V47.8c0-6.4-1.4-9.4-7.9-9.4c-5.4,0-9.2,1.5-11.2,7.3v30.5h-10.6V30.4h10.6v5.8
+		c2.6-4.7,8-6.6,14.4-6.6C307.1,29.6,310.8,35.1,310.8,44.8z"/>
+	<path class="st0" d="M364.8,30.4v42.2c0,12.6-6.5,20.4-24.3,20.4c-4.1,0-8.3-0.4-12.4-1.1v-8.5c3.9,0.9,7.8,1.3,11.8,1.3
+		c10.1,0,14.4-3.3,14.4-12.5v-3.2c-2,4.5-5.7,7.2-14.1,7.2c-15,0-18.7-9.4-18.7-23.3c0-12.6,3.7-23.3,18.7-23.3
+		c8.6,0,12.4,3.1,14.1,7.5v-6.7H364.8z M354.7,52.8c0-9.2-2-15.1-11.9-15.1c-9.4,0-10.4,7.3-10.4,15.1c0,8.5,1.3,15.3,10.4,15.3
+		C352.7,68.2,354.7,62.6,354.7,52.8z"/>
+	<path class="st0" d="M377.3,17.6c0-5.2,1.3-5.7,6-5.7s6,0.5,6,5.7s-1.3,5.8-6,5.8S377.3,22.7,377.3,17.6z M378,30.4h10.6v45.7H378
+		L378,30.4z"/>
+	<path class="st0" d="M442,44.8v31.3h-10.5V47.8c0-6.4-1.4-9.4-7.9-9.4c-5.4,0-9.2,1.5-11.2,7.3v30.5h-10.6V30.4h10.6v5.8
+		c2.6-4.7,8-6.6,14.4-6.6C438.3,29.6,442,35.1,442,44.8z"/>
+	<path class="st0" d="M493.3,55.3l-30,2c0.3,7.7,4,11.2,13.1,11.2c5.1-0.1,10.2-1.1,14.9-3.1V74c-2.7,1.6-9.3,3-16.8,3
+		c-13.9,0-21.9-5.2-21.9-23.7s8-23.7,21.9-23.7c15.8,0,19.1,9,19.1,19.7C493.8,51.3,493.6,53.3,493.3,55.3z M483.9,48.2
+		c0-5.7-1.2-10.5-9.3-10.5c-8.5,0-11,4.1-11.3,12L483.9,48.2z"/>
+	<path class="st1" d="M48.8,11.8c3.1,0,5.6,2.5,5.6,5.6c0,3.1-2.5,5.6-5.6,5.6s-5.6-2.5-5.6-5.6c0,0,0,0,0,0
+		C43.3,14.3,45.7,11.8,48.8,11.8z M11.8,38.5c3.1,0,5.6,2.5,5.6,5.6c0,3.1-2.5,5.6-5.6,5.6s-5.6-2.5-5.6-5.6
+		C6.3,41,8.8,38.5,11.8,38.5z M26.3,82.1c3.1,0,5.6,2.5,5.6,5.6s-2.5,5.6-5.6,5.6c-3.1,0-5.6-2.5-5.6-5.6
+		C20.7,84.6,23.2,82.1,26.3,82.1z M71.3,82.1c3.1,0,5.6,2.5,5.6,5.6s-2.5,5.6-5.6,5.6s-5.6-2.5-5.6-5.6
+		C65.7,84.6,68.2,82.1,71.3,82.1z M57,44.9H40.6l-5,15.5l13.2,9.6L62,60.5L57,44.9L57,44.9z M41.1,43.6L46.8,26
+		c-0.4-0.1-0.9-0.2-1.3-0.4l-5.8,18h-19c0,0.2,0,0.3,0,0.5c0,0.3,0,0.6,0,0.8h18.5l-4.8,14.7l-15-10.9c-0.2,0.4-0.5,0.7-0.8,1.1
+		l12.8,9.3l0,0l2.6,1.9l-0.9,2.5l0,0l-5,15.5c0.4,0.1,0.9,0.2,1.3,0.4l5.7-17.5l12.5,9.1L32.9,81.7c0.3,0.3,0.6,0.7,0.8,1l15.1-11
+		l15.1,11c0.2-0.4,0.5-0.7,0.8-1L49.9,70.9l12.5-9.1l0.6,1.7l0,0l5.2,15.9c0.4-0.2,0.8-0.3,1.3-0.4l-5.8-18l2.6-1.9L79,49.8
+		c-0.3-0.3-0.5-0.7-0.8-1.1l-12.4,9L65.4,58l-2.3,1.6l-4.8-14.7h18.5c0-0.3,0-0.6,0-0.9c0-0.2,0-0.3,0-0.5h-19l-5.8-18
+		c-0.4,0.2-0.8,0.3-1.3,0.4l5.7,17.6L41.1,43.6z M18.9,38.6l22.5-16.4c0.2,0.4,0.5,0.7,0.8,1L19.6,39.7C19.4,39.3,19.1,39,18.9,38.6
+		L18.9,38.6z M22.9,79.4l-8.7-26.8c0.4-0.1,0.8-0.3,1.2-0.4L24.2,79C23.7,79.1,23.3,79.3,22.9,79.4L22.9,79.4z M62.5,88H35.2
+		c0-0.1,0-0.2,0-0.4c0-0.3,0-0.6,0-1h27.3c0,0.3,0,0.6,0,1C62.4,87.8,62.4,87.9,62.5,88L62.5,88z M83.5,52.7l-8.7,26.8
+		c-0.4-0.2-0.8-0.3-1.2-0.4l8.7-26.8C82.6,52.4,83,52.6,83.5,52.7z M56.3,22.2l22.5,16.4c-0.3,0.3-0.5,0.7-0.7,1.1L55.4,23.3
+		C55.7,22.9,56,22.6,56.3,22.2z M85.8,38.5c3.1,0,5.6,2.5,5.6,5.6c0,3.1-2.5,5.6-5.6,5.6s-5.6-2.5-5.6-5.6
+		C80.2,41,82.7,38.5,85.8,38.5z"/>
+</g>
+</svg>
--- a/docs/zh/05-get-started/01-docker.md
+++ b/docs/zh/05-get-started/01-docker.md
@ -10,9 +10,11 @@ title: 通过 Docker 快速体验 TDengine
 如果已经安装了 docker， 只需执行下面的命令。

 ```shell
-docker run -d -p 6030-6049:6030-6049 -p 6030-6049:6030-6049/udp tdengine/tdengine
+docker run -d -p 6030:6030 -p 6041/6041 -p 6043-6049/6043-6049 -p 6043-6049:6043-6049/udp tdengine/tdengine
 ```

+注意：TDengine 3.0 服务端仅使用 6030 TCP 端口。6041 为 taosAdapter 所使用提供 REST 服务端口。6043-6049 为 taosAdapter 提供第三方应用接入所使用端口，可根据需要选择是否打开。
+
 确定该容器已经启动并且在正常运行

 ```shell
@ -51,35 +53,16 @@ $ taos
 Welcome to the TDengine shell from Linux, Client Version:3.0.0.0
 Copyright (c) 2022 by TAOS Data, Inc. All rights reserved.

-Server is Enterprise   trial Edition, ver:3.0.0.0 and will expire at 2022-09-24 15:29:46.
+Server is Community Edition.

 taos> 

 ```

-
-## 启动 REST 服务
-
-taosAdapter 是 TDengine 中提供 REST 服务的组件。下面这条命令会在容器中同时启动 `taosd` 和 `taosadapter` 两个服务组件。
-
-```bash
-docker run -d --name tdengine-all -p 6030-6049:6030-6049 -p 6030-6049:6030-6049/udp tdengine/tdengine
-```
-
-如果想只启动 `taosadapter`：
-
-```bash
-docker run -d --name tdengine-taosa -p 6041-6049:6041-6049 -p 6041-6049:6041-6049/udp -e TAOS_FIRST_EP=tdengine-all tdengine/tdengine:3.0.0.0 taosadapter
-```
-
-如果想只启动 `taosd`：
-
-```bash
-docker run -d --name tdengine-taosd -p 6030-6042:6030-6042 -p 6030-6042:6030-6042/udp -e TAOS_DISABLE_ADAPTER=true tdengine/tdengine:3.0.0.0
-```
-
 ## 访问 REST 接口

+taosAdapter 是 TDengine 中提供 REST 服务的组件。下面这条命令会在容器中同时启动 `taosd` 和 `taosadapter` 两个服务组件。默认 Docker 镜像同时启动 TDengine 后台服务 taosd 和 taosAdatper。
+
 可以在宿主机使用 curl 通过 RESTful 端口访问 Docker 容器内的 TDengine server。

 ```
@ -96,6 +79,22 @@ curl -L -u root:taosdata -d "show databases" 127.0.0.1:6041/rest/sql

 TDengine REST API 详情请参考[官方文档](/reference/rest-api/)。

+## 单独启动 REST 服务
+
+如果想只启动 `taosadapter`：
+
+```bash
+docker run -d --network=host --name tdengine-taosa -e TAOS_FIRST_EP=tdengine-taosd tdengine/tdengine:3.0.0.0 taosadapter
+```
+
+只启动 `taosd`：
+
+```bash
+docker run -d --network=host --name tdengine-taosd -e TAOS_DISABLE_ADAPTER=true tdengine/tdengine:3.0.0.0
+```
+
+注意以上为容器使用 host 方式网络配置进行单独部署 taosAdapter 的命令行参数。其他网络访问方式请设置 hostname、 DNS 等必要的网络配置。
+
 ## 写入数据

 可以使用 TDengine 的自带工具 taosBenchmark 快速体验 TDengine 的写入。
@ -107,7 +106,7 @@ TDengine REST API 详情请参考[官方文档](/reference/rest-api/)。
   
   ```

-   该命令将在数据库 test 下面自动创建一张超级表 meters，该超级表下有 1 万张表，表名为 "d0" 到 "d9999"，每张表有 1 万条记录，每条记录有 (ts, current, voltage, phase) 四个字段，时间戳从 "2017-07-14 10:40:00 000" 到 "2017-07-14 10:40:09 999"，每张表带有标签 location 和 groupId，groupId 被设置为 1 到 10， location 被设置为 "California.SanFrancisco" 或者 "California.LosAngeles"。
+   该命令将在数据库 test 下面自动创建一张超级表 meters，该超级表下有 1 万张表，表名为 "d0" 到 "d9999"，每张表有 1 万条记录，每条记录有 (ts, current, voltage, phase) 四个字段，时间戳从 "2017-07-14 10:40:00 000" 到 "2017-07-14 10:40:09 999"，每张表带有标签 location 和 groupId，groupId 被设置为 1 到 10， location 被设置为 "San Francisco" 或者 "Los Angeles"等城市名称。

   这条命令很快完成 1 亿条记录的插入。具体时间取决于硬件性能。

@ -129,10 +128,10 @@ taos> select count(*) from test.meters;
 taos> select avg(current), max(voltage), min(phase) from test.meters;
 ```

-查询 location="California.SanFrancisco" 的记录总条数：
+查询 location="San Francisco" 的记录总条数：

 ```sql
-taos> select count(*) from test.meters where location="California.SanFrancisco";
+taos> select count(*) from test.meters where location="San Francisco";
 ```

 查询 groupId=10 的所有记录的平均值、最大值、最小值等：
--- a/docs/zh/07-develop/09-udf.md
+++ b/docs/zh/07-develop/09-udf.md
@ -6,56 +6,144 @@ description: "支持用户编码的聚合函数和标量函数，在查询中嵌

 在有些应用场景中，应用逻辑需要的查询无法直接使用系统内置的函数来表示。利用 UDF 功能，TDengine 可以插入用户编写的处理代码并在查询中使用它们，就能够很方便地解决特殊应用场景中的使用需求。 UDF 通常以数据表中的一列数据做为输入，同时支持以嵌套子查询的结果作为输入。

-从 2.2.0.0 版本开始，TDengine 支持通过 C/C++ 语言进行 UDF 定义。接下来结合示例讲解 UDF 的使用方法。
+TDengine 支持通过 C/C++ 语言进行 UDF 定义。接下来结合示例讲解 UDF 的使用方法。

-用户可以通过 UDF 实现两类函数： 标量函数 和 聚合函数。
+用户可以通过 UDF 实现两类函数： 标量函数和聚合函数。标量函数对每行数据返回一个值，如求绝对值 abs，正弦函数 sin，字符串拼接函数 concat 等。聚合函数对多行数据进行返回一个值，如求平均数 avg，最大值 max 等。

-## 用 C/C++ 语言来定义 UDF
+实现 UDF 时，需要实现规定的接口函数
+- 标量函数需要实现标量接口函数 scalarfn 。
+- 聚合函数需要实现聚合接口函数 aggfn_start ， aggfn ， aggfn_finish。
+- 如果需要初始化，实现 udf_init；如果需要清理工作，实现udf_destory。

-### 标量函数
+接口函数的名称是 UDF 名称，或者是 UDF 名称和特定后缀（_start, _finish, _init, _destroy)的连接。列表中的scalarfn，aggfn, udf需要替换成udf函数名。

-用户可以按照下列函数模板定义自己的标量计算函数
+## 实现标量函数
+标量函数实现模板如下
+```c
+#include "taos.h"
+#include "taoserror.h"
+#include "taosudf.h"

- `int32_t udf(SUdfDataBlock* inputDataBlock, SUdfColumn *resultColumn)` 
+// initialization function. if no initialization, we can skip definition of it. The initialization function shall be concatenation of the udf name and _init suffix
+// @return error number defined in taoserror.h
+int32_t scalarfn_init() {
+    // initialization.
+    return TSDB_CODE_SUCCESS;
+}

- 其中 udf 是函数名的占位符，以上述模板实现的函数对行数据块进行标量计算。
+// scalar function main computation function
+// @param inputDataBlock, input data block composed of multiple columns with each column defined by SUdfColumn
+// @param resultColumn, output column
+// @return error number defined in taoserror.h
+int32_t scalarfn(SUdfDataBlock* inputDataBlock, SUdfColumn* resultColumn) {
+    // read data from inputDataBlock and process, then output to resultColumn.
+    return TSDB_CODE_SUCCESS;
+}

- scalarFunction 中各参数的具体含义是：
+// cleanup function. if no cleanup related processing, we can skip definition of it. The destroy function shall be concatenation of the udf name and _destroy suffix.
+// @return error number defined in taoserror.h
+int32_t scalarfn_destroy() {
+    // clean up
+    return TSDB_CODE_SUCCESS;
+}
+```
+scalarfn 为函数名的占位符，需要替换成函数名，如bit_and。
+
+## 实现聚合函数
+
+聚合函数的实现模板如下
+```c
+#include "taos.h"
+#include "taoserror.h"
+#include "taosudf.h"
+
+// Initialization function. if no initialization, we can skip definition of it. The initialization function shall be concatenation of the udf name and _init suffix
+// @return error number defined in taoserror.h
+int32_t aggfn_init() {
+    // initialization.
+    return TSDB_CODE_SUCCESS;
+}
+
+// aggregate start function. The intermediate value or the state(@interBuf) is initialized in this function. The function name shall be concatenation of udf name and _start suffix
+// @param interbuf intermediate value to intialize
+// @return error number defined in taoserror.h
+int32_t aggfn_start(SUdfInterBuf* interBuf) {
+    // initialize intermediate value in interBuf
+    return TSDB_CODE_SUCESS;
+}
+
+// aggregate reduce function. This function aggregate old state(@interbuf) and one data bock(inputBlock) and output a new state(@newInterBuf).
+// @param inputBlock input data block
+// @param interBuf old state
+// @param newInterBuf new state
+// @return error number defined in taoserror.h
+int32_t aggfn(SUdfDataBlock* inputBlock, SUdfInterBuf *interBuf, SUdfInterBuf *newInterBuf) {
+    // read from inputBlock and interBuf and output to newInterBuf
+    return TSDB_CODE_SUCCESS;
+}
+
+// aggregate function finish function. This function transforms the intermediate value(@interBuf) into the final output(@result). The function name must be concatenation of aggfn and _finish suffix.
+// @interBuf : intermediate value
+// @result: final result
+// @return error number defined in taoserror.h
+int32_t int32_t aggfn_finish(SUdfInterBuf* interBuf, SUdfInterBuf *result) {
+    // read data from inputDataBlock and process, then output to result
+    return TSDB_CODE_SUCCESS;
+}
+
+// cleanup function. if no cleanup related processing, we can skip definition of it. The destroy function shall be concatenation of the udf name and _destroy suffix.
+// @return error number defined in taoserror.h
+int32_t aggfn_destroy() {
+    // clean up
+    return TSDB_CODE_SUCCESS;
+}
+```
+aggfn为函数名的占位符，需要修改为自己的函数名，如l2norm。
+
+## 接口函数定义
+
+接口函数的名称是 udf 名称，或者是 udf 名称和特定后缀（_start, _finish, _init, _destroy)的连接。以下描述中函数名称中的 scalarfn，aggfn, udf 需要替换成udf函数名。
+
+接口函数返回值表示是否成功，如果错误返回错误代码，错误代码见taoserror.h。
+
+接口函数参数类型见数据结构定义。
+
+### 标量接口函数
+
+ `int32_t scalarfn(SUdfDataBlock* inputDataBlock, SUdfColumn *resultColumn)` 
+ 
+ 其中 scalarFn 是函数名的占位符。这个函数对数据块进行标量计算，通过设置resultColumn结构体中的变量设置值
+
+参数的具体含义是：
  - inputDataBlock: 输入的数据块
-  - resultColumn: 输出列 
+  - resultColumn: 输出列。输出列 

-### 聚合函数
+### 聚合接口函数

-用户可以按照如下函数模板定义自己的聚合函数。
+`int32_t aggfn_start(SUdfInterBuf *interBuf)`

-`int32_t udf_start(SUdfInterBuf *interBuf)`
+`int32_t aggfn(SUdfDataBlock* inputBlock, SUdfInterBuf *interBuf, SUdfInterBuf *newInterBuf)`

-`int32_t udf(SUdfDataBlock* inputBlock, SUdfInterBuf *interBuf, SUdfInterBuf *newInterBuf)`
+`int32_t aggfn_finish(SUdfInterBuf* interBuf, SUdfInterBuf *result)`

-`int32_t udf_finish(SUdfInterBuf* interBuf, SUdfInterBuf *result)`
-其中 udf 是函数名的占位符。其中各参数的具体含义是：
+其中 aggfn 是函数名的占位符。首先调用aggfn_start生成结果buffer，然后相关的数据会被分为多个行数据块，对每个数据块调用 aggfn 用数据块更新中间结果，最后再调用 aggfn_finish 从中间结果产生最终结果，最终结果只能含 0 或 1 条结果数据。

+参数的具体含义是：
  - interBuf：中间结果 buffer。
  - inputBlock：输入的数据块。
  - newInterBuf：新的中间结果buffer。
  - result：最终结果。


-其计算过程为：首先调用udf_start生成结果buffer，然后相关的数据会被分为多个行数据块，对每个行数据块调用 udf 用数据块更新中间结果，最后再调用 udf_finish 从中间结果产生最终结果，最终结果只能含 0 或 1 条结果数据。
-
 ### UDF 初始化和销毁
 `int32_t udf_init()`

 `int32_t udf_destroy()`

-其中 udf 是函数名的占位符。udf_init 完成初始化工作。 udf_destroy 完成清理工作。
+其中 udf 是函数名的占位符。udf_init 完成初始化工作。 udf_destroy 完成清理工作。如果没有初始化工作，无需定义udf_init函数。如果没有清理工作，无需定义udf_destroy函数。

-:::note
-如果对应的函数不需要具体的功能，也需要实现一个空函数。

-:::
-
-### UDF 数据结构
+## UDF 数据结构
 ```c
 typedef struct SUdfColumnMeta {
  int16_t type;
@ -103,6 +191,13 @@ typedef struct SUdfInterBuf {
  int8_t numOfResult; //zero or one
 } SUdfInterBuf;
 ```
+数据结构说明如下：
+
+- SUdfDataBlock 数据块包含行数 numOfRows 和列数 numCols。udfCols[i] (0 <= i <= numCols-1)表示每一列数据，类型为SUdfColumn*。
+- SUdfColumn 包含列的数据类型定义 colMeta 和列的数据 colData。
+- SUdfColumnMeta 成员定义同 taos.h 数据类型定义。
+- SUdfColumnData 数据可以变长，varLenCol 定义变长数据，fixLenCol 定义定长数据。 
+- SUdfInterBuf 定义中间结构 buffer，以及 buffer 中结果个数 numOfResult

 为了更好的操作以上数据结构，提供了一些便利函数，定义在 taosudf.h。

@ -118,75 +213,15 @@ gcc -g -O0 -fPIC -shared add_one.c -o add_one.so

 这样就准备好了动态链接库 add_one.so 文件，可以供后文创建 UDF 时使用了。为了保证可靠的系统运行，编译器 GCC 推荐使用 7.5 及以上版本。

-## 在系统中管理和使用 UDF
-
-### 创建 UDF
-
-用户可以通过 SQL 指令在系统中加载客户端所在主机上的 UDF 函数库（不能通过 RESTful 接口或 HTTP 管理界面来进行这一过程）。一旦创建成功，则当前 TDengine 集群的所有用户都可以在 SQL 指令中使用这些函数。UDF 存储在系统的 MNode 节点上，因此即使重启 TDengine 系统，已经创建的 UDF 也仍然可用。
-
-在创建 UDF 时，需要区分标量函数和聚合函数。如果创建时声明了错误的函数类别，则可能导致通过 SQL 指令调用函数时出错。此外，用户需要保证输入数据类型与 UDF 程序匹配，UDF 输出数据类型与 OUTPUTTYPE 匹配。
-
- 创建标量函数
-```sql
-CREATE FUNCTION function_name AS library_path OUTPUTTYPE output_type;
-```
-
-  - function_name：标量函数未来在 SQL 中被调用时的函数名，必须与函数实现中 udf 的实际名称一致；
-  - library_path：包含 UDF 函数实现的动态链接库的库文件绝对路径（指的是库文件在当前客户端所在主机上的保存路径，通常是指向一个 .so 文件），这个路径需要用英文单引号或英文双引号括起来；
-  - output_type：此函数计算结果的数据类型名称；
-
-  例如，如下语句可以把 libbitand.so 创建为系统中可用的 UDF：
-
-  ```sql
-  CREATE FUNCTION bit_and AS "/home/taos/udf_example/libbitand.so" OUTPUTTYPE INT;
-  ```
-
- 创建聚合函数：
-```sql
-CREATE AGGREGATE FUNCTION function_name AS library_path OUTPUTTYPE output_type [ BUFSIZE buffer_size ];
-```
-
-  - function_name：聚合函数未来在 SQL 中被调用时的函数名，必须与函数实现中 udfNormalFunc 的实际名称一致；
-  - library_path：包含 UDF 函数实现的动态链接库的库文件绝对路径（指的是库文件在当前客户端所在主机上的保存路径，通常是指向一个 .so 文件），这个路径需要用英文单引号或英文双引号括起来；
-  - output_type：此函数计算结果的数据类型，与上文中 udfNormalFunc 的 itype 参数不同，这里不是使用数字表示法，而是直接写类型名称即可；
-  - buffer_size：中间计算结果的缓冲区大小，单位是字节。如果不使用可以不设置。
-
-  例如，如下语句可以把 libsqrsum.so 创建为系统中可用的 UDF：
-
-  ```sql
-  CREATE AGGREGATE FUNCTION sqr_sum AS "/home/taos/udf_example/libsqrsum.so" OUTPUTTYPE DOUBLE bufsize 8;
-  ```
-
-### 管理 UDF
-
- 删除指定名称的用户定义函数：
-```
-DROP FUNCTION function_name;
-```
-
- function_name：此参数的含义与 CREATE 指令中的 function_name 参数一致，也即要删除的函数的名字，例如 
-```sql
-DROP FUNCTION bit_and;
-```
- 显示系统中当前可用的所有 UDF：
-```sql
-SHOW FUNCTIONS;
-```
-
-### 调用 UDF
-
-在 SQL 指令中，可以直接以在系统中创建 UDF 时赋予的函数名来调用用户定义函数。例如：
-```sql
-SELECT X(c1,c2) FROM table/stable;
-```
-
-表示对名为 c1, c2 的数据列调用名为 X 的用户定义函数。SQL 指令中用户定义函数可以配合 WHERE 等查询特性来使用。
-
+## 管理和使用UDF
+关于如何管理和使用UDF，参见[UDF使用说明](../12-taos-sql/26-udf.md)

 ## 示例代码

 ### 标量函数示例 [bit_and](https://github.com/taosdata/TDengine/blob/develop/tests/script/sh/bit_and.c)

+bit_add 实现多列的按位与功能。如果只有一列，返回这一列。bit_add 忽略空值。
+
 <details>
 <summary>bit_and.c</summary>

@ -196,13 +231,15 @@ SELECT X(c1,c2) FROM table/stable;

 </details>

-### 聚合函数示例 [sqr_sum](https://github.com/taosdata/TDengine/blob/develop/tests/script/sh/sqr_sum.c)
+### 聚合函数示例 [l2norm](https://github.com/taosdata/TDengine/blob/develop/tests/script/sh/l2norm.c)
+
+l2norm 实现了输入列的所有数据的二阶范数，即对每个数据先平方，再累加求和，最后开方。

 <details>
-<summary>sqr_sum.c</summary>
+<summary>l2norm.c</summary>

 ```c
-{{#include tests/script/sh/sqr_sum.c}}
+{{#include tests/script/sh/l2norm.c}}
 ```

 </details>
--- a/docs/zh/12-taos-sql/26-udf.md
+++ b/docs/zh/12-taos-sql/26-udf.md
@ -4,34 +4,65 @@ title: 用户自定义函数
 ---

 除了 TDengine 的内置函数以外，用户还可以编写自己的函数逻辑并加入TDengine系统中。
+## 创建 UDF

-## 创建函数
+用户可以通过 SQL 指令在系统中加载客户端所在主机上的 UDF 函数库（不能通过 RESTful 接口或 HTTP 管理界面来进行这一过程）。一旦创建成功，则当前 TDengine 集群的所有用户都可以在 SQL 指令中使用这些函数。UDF 存储在系统的 MNode 节点上，因此即使重启 TDengine 系统，已经创建的 UDF 也仍然可用。

+在创建 UDF 时，需要区分标量函数和聚合函数。如果创建时声明了错误的函数类别，则可能导致通过 SQL 指令调用函数时出错。此外，用户需要保证输入数据类型与 UDF 程序匹配，UDF 输出数据类型与 OUTPUTTYPE 匹配。
+
+- 创建标量函数
 ```sql
-CREATE [AGGREGATE] FUNCTION func_name AS library_path OUTPUTTYPE type_name [BUFSIZE buffer_size]
+CREATE FUNCTION function_name AS library_path OUTPUTTYPE output_type;
 ```

-语法说明：
+  - function_name：标量函数未来在 SQL 中被调用时的函数名，必须与函数实现中 udf 的实际名称一致；
+  - library_path：包含 UDF 函数实现的动态链接库的库文件绝对路径（指的是库文件在当前客户端所在主机上的保存路径，通常是指向一个 .so 文件），这个路径需要用英文单引号或英文双引号括起来；
+  - output_type：此函数计算结果的数据类型名称；

-AGGREGATE：标识此函数是标量函数还是聚集函数。
-func_name：函数名，必须与函数实现中 udf 的实际名称一致。
-library_path：包含UDF函数实现的动态链接库的绝对路径，是在客户端侧主机上的绝对路径。
-type_name：标识此函数的返回类型。
-buffer_size：中间结果的缓冲区大小，单位是字节。不设置则默认为0。
+  例如，如下语句可以把 libbitand.so 创建为系统中可用的 UDF：

+  ```sql
+  CREATE FUNCTION bit_and AS "/home/taos/udf_example/libbitand.so" OUTPUTTYPE INT;
+  ```
+
+- 创建聚合函数：
+```sql
+CREATE AGGREGATE FUNCTION function_name AS library_path OUTPUTTYPE output_type [ BUFSIZE buffer_size ];
+```
+
+  - function_name：聚合函数未来在 SQL 中被调用时的函数名，必须与函数实现中 udfNormalFunc 的实际名称一致；
+  - library_path：包含 UDF 函数实现的动态链接库的库文件绝对路径（指的是库文件在当前客户端所在主机上的保存路径，通常是指向一个 .so 文件），这个路径需要用英文单引号或英文双引号括起来；
+  - output_type：此函数计算结果的数据类型，与上文中 udfNormalFunc 的 itype 参数不同，这里不是使用数字表示法，而是直接写类型名称即可；
+  - buffer_size：中间计算结果的缓冲区大小，单位是字节。如果不使用可以不设置。
+
+  例如，如下语句可以把 libl2norm.so 创建为系统中可用的 UDF：
+
+  ```sql
+  CREATE AGGREGATE FUNCTION l2norm AS "/home/taos/udf_example/libl2norm.so" OUTPUTTYPE DOUBLE bufsize 8;
+  ```
 关于如何开发自定义函数，请参考 [UDF使用说明](../../develop/udf)。

-## 删除自定义函数
+## 管理 UDF

+- 删除指定名称的用户定义函数：
 ```
 DROP FUNCTION function_name;
 ```

- function_name：此参数的含义与 CREATE 指令中的 function_name 参数一致，也即要删除的函数的名字，例如 
-
-
-## 显示 UDF
-
+- function_name：此参数的含义与 CREATE 指令中的 function_name 参数一致，也即要删除的函数的名字，例如bit_and, l2norm 
 ```sql
-SHOW FUNCTION;
+DROP FUNCTION bit_and;
 ```
+- 显示系统中当前可用的所有 UDF：
+```sql
+SHOW FUNCTIONS;
+```
+
+## 调用 UDF
+
+在 SQL 指令中，可以直接以在系统中创建 UDF 时赋予的函数名来调用用户定义函数。例如：
+```sql
+SELECT X(c1,c2) FROM table/stable;
+```
+
+表示对名为 c1, c2 的数据列调用名为 X 的用户定义函数。SQL 指令中用户定义函数可以配合 WHERE 等查询特性来使用。
--- a/docs/zh/12-taos-sql/index.md
+++ b/docs/zh/12-taos-sql/index.md
@ -5,11 +5,12 @@ description: "TAOS SQL 支持的语法规则、主要查询功能、支持的 SQ

 本文档说明 TAOS SQL 支持的语法规则、主要查询功能、支持的 SQL 查询函数，以及常用技巧等内容。阅读本文档需要读者具有基本的 SQL 语言的基础。

-TAOS SQL 是用户对 TDengine 进行数据写入和查询的主要工具。TAOS SQL 为了便于用户快速上手，在一定程度上提供与标准 SQL 类似的风格和模式。严格意义上，TAOS SQL 并不是也不试图提供标准的 SQL 语法。此外，由于 TDengine 针对的时序性结构化数据不提供删除功能，因此在 TAO SQL 中不提供数据删除的相关功能。
+TAOS SQL 是用户对 TDengine 进行数据写入和查询的主要工具。TAOS SQL 提供标准的 SQL 语法，并针对时序数据和业务的特点优化和新增了许多语法和功能。TAOS SQL 语句的最大长度为 1M。TAOS SQL 不支持关键字的缩写，例如 DELETE 不能缩写为 DEL。

 本章节 SQL 语法遵循如下约定：

- <\> 里的内容是用户需要输入的，但不要输入 <\> 本身
+- 用大写字母表示关键字，但 SQL 本身并不区分关键字和标识符的大小写
+- 用小写字母表示需要用户输入的内容
 - \[ \] 表示内容为可选项，但不能输入 [] 本身
 - | 表示多选一，选择其中一个即可，但不能输入 | 本身
 - … 表示前面的项可重复多个
--- a/docs/zh/14-reference/03-connector/java.mdx
+++ b/docs/zh/14-reference/03-connector/java.mdx
@ -9,7 +9,7 @@ description: TDengine Java 连接器基于标准 JDBC API 实现, 并提供原
 import Tabs from '@theme/Tabs';
 import TabItem from '@theme/TabItem';

-`taos-jdbcdriver` 是 TDengine 的官方 Java 语言连接器，Java 开发人员可以通过它开发存取 TDengine 数据库的应用软件。`taos-jdbcdriver` 实现了 JDBC driver 标准的接口，并提供两种形式的连接器。一种是通过 TDengine 客户端驱动程序（taosc）原生连接 TDengine 实例，支持数据写入、查询、订阅、schemaless 接口和参数绑定接口等功能，一种是通过 taosAdapter 提供的 REST 接口连接 TDengine 实例（2.4.0.0 及更高版本）。REST 连接实现的功能集合和原生连接有少量不同。
+`taos-jdbcdriver` 是 TDengine 的官方 Java 语言连接器，Java 开发人员可以通过它开发存取 TDengine 数据库的应用软件。`taos-jdbcdriver` 实现了 JDBC driver 标准的接口，并提供两种形式的连接器。一种是通过 TDengine 客户端驱动程序（taosc）原生连接 TDengine 实例，支持数据写入、查询、订阅、schemaless 接口和参数绑定接口等功能，一种是通过 taosAdapter 提供的 REST 接口连接 TDengine 实例。REST 连接实现的功能集合和原生连接有少量不同。

 ![TDengine Database Connector Java](tdengine-jdbc-connector.webp)

@ -41,19 +41,19 @@ REST 连接支持所有能运行 Java 的平台。

 TDengine 目前支持时间戳、数字、字符、布尔类型，与 Java 对应类型转换如下：

-| TDengine DataType | JDBCType （driver 版本 < 2.0.24） | JDBCType （driver 版本 >= 2.0.24） |
-| ----------------- | --------------------------------- | ---------------------------------- |
-| TIMESTAMP         | java.lang.Long                    | java.sql.Timestamp                 |
-| INT               | java.lang.Integer                 | java.lang.Integer                  |
-| BIGINT            | java.lang.Long                    | java.lang.Long                     |
-| FLOAT             | java.lang.Float                   | java.lang.Float                    |
-| DOUBLE            | java.lang.Double                  | java.lang.Double                   |
-| SMALLINT          | java.lang.Short                   | java.lang.Short                    |
-| TINYINT           | java.lang.Byte                    | java.lang.Byte                     |
-| BOOL              | java.lang.Boolean                 | java.lang.Boolean                  |
-| BINARY            | java.lang.String                  | byte array                         |
-| NCHAR             | java.lang.String                  | java.lang.String                   |
-| JSON              | -                                 | java.lang.String                   |
+| TDengine DataType | JDBCType  |
+| ----------------- | ---------------------------------- |
+| TIMESTAMP         | java.sql.Timestamp                 |
+| INT               | java.lang.Integer                  |
+| BIGINT            | java.lang.Long                     |
+| FLOAT             | java.lang.Float                    |
+| DOUBLE            | java.lang.Double                   |
+| SMALLINT          | java.lang.Short                    |
+| TINYINT           | java.lang.Byte                     |
+| BOOL              | java.lang.Boolean                  |
+| BINARY            | byte array                         |
+| NCHAR             | java.lang.String                   |
+| JSON              | java.lang.String                   |

 **注意**：JSON 类型仅在 tag 中支持。

@ -198,7 +198,7 @@ url 中的配置参数如下：

 - user：登录 TDengine 用户名，默认值 'root'。
 - password：用户登录密码，默认值 'taosdata'。
- batchfetch: true：在执行查询时批量拉取结果集；false：逐行拉取结果集。默认值为：false。逐行拉取结果集使用 HTTP 方式进行数据传输。从 taos-jdbcdriver-2.0.38 和 TDengine 2.4.0.12 版本开始，JDBC REST 连接增加批量拉取数据功能。taos-jdbcdriver 与 TDengine 之间通过 WebSocket 连接进行数据传输。相较于 HTTP，WebSocket 可以使 JDBC REST 连接支持大数据量查询，并提升查询性能。
+- batchfetch: true：在执行查询时批量拉取结果集；false：逐行拉取结果集。默认值为：false。逐行拉取结果集使用 HTTP 方式进行数据传输。从 taos-jdbcdriver-2.0.38 开始，JDBC REST 连接增加批量拉取数据功能。taos-jdbcdriver 与 TDengine 之间通过 WebSocket 连接进行数据传输。相较于 HTTP，WebSocket 可以使 JDBC REST 连接支持大数据量查询，并提升查询性能。
 - charset: 当开启批量拉取数据时，指定解析字符串数据的字符集。
 - batchErrorIgnore：true：在执行 Statement 的 executeBatch 时，如果中间有一条 SQL 执行失败，继续执行下面的 SQL 了。false：不再执行失败 SQL 后的任何语句。默认值为：false。
 - httpConnectTimeout: 连接超时时间，单位 ms， 默认值为 5000。
@ -216,7 +216,7 @@ url 中的配置参数如下：
 INSERT INTO test.t1 USING test.weather (ts, temperature) TAGS('California.SanFrancisco') VALUES(now, 24.6);
 ```

- 从 taos-jdbcdriver-2.0.36 和 TDengine 2.2.0.0 版本开始，如果在 url 中指定了 dbname，那么，JDBC REST 连接会默认使用/rest/sql/dbname 作为 restful 请求的 url，在 SQL 中不需要指定 dbname。例如：url 为 jdbc:TAOS-RS://127.0.0.1:6041/test，那么，可以执行 sql：insert into t1 using weather(ts, temperature) tags('California.SanFrancisco') values(now, 24.6);
+- 从 taos-jdbcdriver-2.0.36 开始，如果在 url 中指定了 dbname，那么，JDBC REST 连接会默认使用/rest/sql/dbname 作为 restful 请求的 url，在 SQL 中不需要指定 dbname。例如：url 为 jdbc:TAOS-RS://127.0.0.1:6041/test，那么，可以执行 sql：insert into t1 using weather(ts, temperature) tags('California.SanFrancisco') values(now, 24.6);

 :::

@ -358,11 +358,11 @@ JDBC 连接器可能报错的错误码包括 3 种：JDBC driver 本身的报错
 具体的错误码请参考：

 - [TDengine Java Connector](https://github.com/taosdata/taos-connector-jdbc/blob/main/src/main/java/com/taosdata/jdbc/TSDBErrorNumbers.java)
- [TDengine_ERROR_CODE](https://github.com/taosdata/TDengine/blob/develop/src/inc/taoserror.h)
+- [TDengine_ERROR_CODE](../error-code)

 ### 通过参数绑定写入数据

-从 2.1.2.0 版本开始，TDengine 的 JDBC 原生连接实现大幅改进了参数绑定方式对数据写入（INSERT）场景的支持。采用这种方式写入数据时，能避免 SQL 语法解析的资源消耗，从而在很多情况下显著提升写入性能。
+TDengine 的 JDBC 原生连接实现大幅改进了参数绑定方式对数据写入（INSERT）场景的支持。采用这种方式写入数据时，能避免 SQL 语法解析的资源消耗，从而在很多情况下显著提升写入性能。

 **注意**：

@ -630,7 +630,7 @@ public void setNString(int columnIndex, ArrayList<String> list, int size) throws

 ### 无模式写入

-从 2.2.0.0 版本开始，TDengine 增加了对无模式写入功能。无模式写入兼容 InfluxDB 的 行协议（Line Protocol）、OpenTSDB 的 telnet 行协议和 OpenTSDB 的 JSON 格式协议。详情请参见[无模式写入](/reference/schemaless/)。
+TDengine 支持无模式写入功能。无模式写入兼容 InfluxDB 的 行协议（Line Protocol）、OpenTSDB 的 telnet 行协议和 OpenTSDB 的 JSON 格式协议。详情请参见[无模式写入](../../schemaless)。

 **注意**：

@ -670,55 +670,127 @@ public class SchemalessInsertTest {

 TDengine Java 连接器支持订阅功能，应用 API 如下：

-#### 创建订阅
+#### 创建 Topic

 ```java
-TSDBSubscribe sub = ((TSDBConnection)conn).subscribe("topic", "select * from meters", false);
+Connection connection = DriverManager.getConnection(url, properties);
+Statement statement = connection.createStatement();
+statement.executeUpdate("create topic if not exists topic_speed as select ts, speed from speed_table");
 ```

 `subscribe` 方法的三个参数含义如下：

- topic：订阅的主题（即名称），此参数是订阅的唯一标识
- sql：订阅的查询语句，此语句只能是 `select` 语句，只应查询原始数据，只能按时间正序查询数据
- restart：如果订阅已经存在，是重新开始，还是继续之前的订阅
+- topic_speed：订阅的主题（即名称），此参数是订阅的唯一标识。
+- sql：订阅的查询语句，此语句只能是 `select` 语句，只应查询原始数据，只能按时间正序查询数据。

-如上面的例子将使用 SQL 语句 `select * from meters` 创建一个名为 `topic` 的订阅，如果这个订阅已经存在，将继续之前的查询进度，而不是从头开始消费所有的数据。
+如上面的例子将使用 SQL 语句 `select ts, speed from speed_table` 创建一个名为 `topic_speed` 的订阅。
+
+#### 创建 Consumer
+
+```java
+Properties config = new Properties();
+config.setProperty("enable.auto.commit", "true");
+config.setProperty("group.id", "group1");
+config.setProperty("value.deserializer", "com.taosdata.jdbc.tmq.ConsumerTest.ResultDeserializer");
+
+TaosConsumer consumer = new TaosConsumer<>(config);
+```
+
+- enable.auto.commit: 是否允许自动提交。
+- group.id: consumer: 所在的 group。
+- value.deserializer: 结果集反序列化方法，可以继承 `com.taosdata.jdbc.tmq.ReferenceDeserializer`，并指定结果集 bean，实现反序列化。也可以继承 `com.taosdata.jdbc.tmq.Deserializer`，根据 SQL 的 resultSet 自定义反序列化方式。
+- 其他参数请参考：[Consumer 参数列表](../../../develop/tmq#创建-consumer-以及consumer-group)

 #### 订阅消费数据

 ```java
-int total = 0;
 while(true) {
-    TSDBResultSet rs = sub.consume();
-    int count = 0;
-    while(rs.next()) {
-        count++;
+    ConsumerRecords<ResultBean> records = consumer.poll(Duration.ofMillis(100));
+        for (ResultBean record : records) {
+            process(record);
        }
-    total += count;
-    System.out.printf("%d rows consumed, total %d\n", count, total);
-    Thread.sleep(1000);
 }
 ```

-`consume` 方法返回一个结果集，其中包含从上次 `consume` 到目前为止的所有新数据。请务必按需选择合理的调用 `consume` 的频率（如例子中的 `Thread.sleep(1000)`），否则会给服务端造成不必要的压力。
+`poll` 方法返回一个结果集，其中包含从上次 `poll` 到目前为止的所有新数据。请务必按需选择合理的调用 `poll` 的频率（如例子中的 `Duration.ofMillis(100)`），否则会给服务端造成不必要的压力。

 #### 关闭订阅

 ```java
-sub.close(true);
+consumer.close()
 ```

-`close` 方法关闭一个订阅。如果其参数为 `true` 表示保留订阅进度信息，后续可以创建同名订阅继续消费数据；如为 `false` 则不保留订阅进度。
-
-### 关闭资源
+### 使用示例如下：

 ```java
-resultSet.close();
-stmt.close();
-conn.close();
-```
+public abstract class ConsumerLoop {
+    private final TaosConsumer<ResultBean> consumer;
+    private final List<String> topics;
+    private final AtomicBoolean shutdown;
+    private final CountDownLatch shutdownLatch;

-> `注意务必要将 connection 进行关闭`，否则会出现连接泄露。
+    public ConsumerLoop() throws SQLException {
+        Properties config = new Properties();
+        config.setProperty("msg.with.table.name", "true");
+        config.setProperty("enable.auto.commit", "true");
+        config.setProperty("group.id", "group1");
+        config.setProperty("value.deserializer", "com.taosdata.jdbc.tmq.ConsumerTest.ResultDeserializer");
+
+        this.consumer = new TaosConsumer<>(config);
+        this.topics = Collections.singletonList("topic_speed");
+        this.shutdown = new AtomicBoolean(false);
+        this.shutdownLatch = new CountDownLatch(1);
+    }
+
+    public abstract void process(ResultBean result);
+
+    public void pollData() throws SQLException {
+        try {
+            consumer.subscribe(topics);
+
+            while (!shutdown.get()) {
+                ConsumerRecords<ResultBean> records = consumer.poll(Duration.ofMillis(100));
+                for (ResultBean record : records) {
+                    process(record);
+                }
+            }
+        } finally {
+             consumer.close();
+            shutdownLatch.countDown();
+        }
+    }
+
+    public void shutdown() throws InterruptedException {
+        shutdown.set(true);
+        shutdownLatch.await();
+    }
+
+    static class ResultDeserializer extends ReferenceDeserializer<ResultBean> {
+
+    }
+
+    static class ResultBean {
+        private Timestamp ts;
+        private int speed;
+
+        public Timestamp getTs() {
+            return ts;
+        }
+
+        public void setTs(Timestamp ts) {
+            this.ts = ts;
+        }
+
+        public int getSpeed() {
+            return speed;
+        }
+
+        public void setSpeed(int speed) {
+            this.speed = speed;
+        }
+    }
+}
+```

 ### 与连接池使用

@ -787,20 +859,6 @@ public static void main(String[] args) throws Exception {

 > 更多 druid 使用问题请查看[官方说明](https://github.com/alibaba/druid)。

-**注意事项：**
-
- TDengine `v1.6.4.1` 版本开始提供了一个专门用于心跳检测的函数 `select server_status()`，所以在使用连接池时推荐使用 `select server_status()` 进行 Validation Query。
-
-如下所示，`select server_status()` 执行成功会返回 `1`。
-
-```sql
-taos> select server_status();
-server_status()|
-================
-1              |
-Query OK, 1 row(s) in set (0.000141s)
-```
-
 ### 更多示例程序

 示例程序源码位于 `TDengine/examples/JDBC` 下:
@ -811,7 +869,7 @@ Query OK, 1 row(s) in set (0.000141s)
 - SpringJdbcTemplate：Spring JdbcTemplate 中使用 taos-jdbcdriver。
 - mybatisplus-demo：Springboot + Mybatis 中使用 taos-jdbcdriver。

-请参考：[JDBC example](https://github.com/taosdata/TDengine/tree/develop/examples/JDBC)
+请参考：[JDBC example](https://github.com/taosdata/TDengine/tree/3.0/examples/JDBC)

 ## 最近更新记录

@ -842,7 +900,7 @@ Query OK, 1 row(s) in set (0.000141s)

   **解决方法**：重新安装 64 位 JDK。

-4. 其它问题请参考 [FAQ](/train-faq/faq)
+4. 其它问题请参考 [FAQ](../../../train-faq/faq)

 ## API 参考

--- a/examples/rust
+++ b/examples/rust
@ -0,0 +1 @@
+Subproject commit 7ed7a97715388fa144718764d6bf20f9bfc29a12
--- a/include/libs/stream/tstream.h
+++ b/include/libs/stream/tstream.h
@ -42,8 +42,8 @@ enum {
  TASK_STATUS__DROPPING,
  TASK_STATUS__FAIL,
  TASK_STATUS__STOP,
-  TASK_STATUS__PREPARE_RECOVER,
-  TASK_STATUS__RECOVERING,
+  TASK_STATUS__RECOVER_DOWNSTREAM,
+  TASK_STATUS__RECOVER_SELF,
 };

 enum {
@ -226,11 +226,36 @@ typedef struct {
  int32_t nodeId;
  int32_t childId;
  int32_t taskId;
-  int64_t checkpointVer;
-  int64_t processedVer;
+  // int64_t checkpointVer;
+  // int64_t processedVer;
  SEpSet epSet;
 } SStreamChildEpInfo;

+typedef struct {
+  int32_t srcNodeId;
+  int32_t srcChildId;
+  int64_t stateSaveVer;
+  int64_t stateProcessedVer;
+} SStreamCheckpointInfo;
+
+typedef struct {
+  int64_t streamId;
+  int64_t checkTs;
+  int32_t checkpointId;  // incremental
+  int32_t taskId;
+  SArray* checkpointVer;  // SArray<SStreamCheckpointInfo>
+} SStreamMultiVgCheckpointInfo;
+
+typedef struct {
+  int32_t taskId;
+  int32_t checkpointId;  // incremental
+} SStreamCheckpointKey;
+
+typedef struct {
+  int32_t taskId;
+  SArray* checkpointVer;
+} SStreamRecoveringState;
+
 typedef struct SStreamTask {
  int64_t streamId;
  int32_t taskId;
@ -256,6 +281,8 @@ typedef struct SStreamTask {

  // children info
  SArray* childEpInfo;  // SArray<SStreamChildEpInfo*>
+  int32_t nextCheckId;
+  SArray* checkpointInfo;  // SArray<SStreamCheckpointInfo>

  // exec
  STaskExec exec;
@ -367,9 +394,6 @@ typedef struct {
  int32_t upstreamTaskId;
  int32_t upstreamChildId;
  int32_t upstreamNodeId;
-#if 0
-  int64_t sourceVer;
-#endif
  int32_t blockNum;
  SArray* dataLen;  // SArray<int32_t>
  SArray* data;     // SArray<SRetrieveTableRsp*>
@ -399,6 +423,7 @@ typedef struct {
  int32_t rspToTaskId;
 } SStreamRetrieveRsp;

+#if 0
 typedef struct {
  int64_t streamId;
  int32_t taskId;
@ -435,24 +460,35 @@ int32_t tDecodeSMStreamTaskRecoverReq(SDecoder* pDecoder, SMStreamTaskRecoverReq
 int32_t tEncodeSMStreamTaskRecoverRsp(SEncoder* pEncoder, const SMStreamTaskRecoverRsp* pRsp);
 int32_t tDecodeSMStreamTaskRecoverRsp(SDecoder* pDecoder, SMStreamTaskRecoverRsp* pRsp);

-typedef struct {
-  int64_t streamId;
-} SPStreamTaskRecoverReq;
+int32_t streamProcessRecoverReq(SStreamTask* pTask, SStreamTaskRecoverReq* pReq, SRpcMsg* pMsg);
+int32_t streamProcessRecoverRsp(SStreamTask* pTask, SStreamTaskRecoverRsp* pRsp);
+#endif

 typedef struct {
-  int8_t reserved;
-} SPStreamTaskRecoverRsp;
+  int64_t streamId;
+  int32_t downstreamTaskId;
+  int32_t taskId;
+} SStreamRecoverDownstreamReq;
+
+typedef struct {
+  int64_t streamId;
+  int32_t downstreamTaskId;
+  int32_t taskId;
+  SArray* checkpointVer;  // SArray<SStreamCheckpointInfo>
+} SStreamRecoverDownstreamRsp;
+
+int32_t tEncodeSStreamTaskRecoverReq(SEncoder* pEncoder, const SStreamRecoverDownstreamReq* pReq);
+int32_t tDecodeSStreamTaskRecoverRsp(SDecoder* pDecoder, const SStreamRecoverDownstreamRsp* pRsp);

 int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq);
 int32_t tDecodeStreamRetrieveReq(SDecoder* pDecoder, SStreamRetrieveReq* pReq);
+void    tFreeStreamDispatchReq(SStreamDispatchReq* pReq);

 int32_t streamSetupTrigger(SStreamTask* pTask);

 int32_t streamProcessRunReq(SStreamTask* pTask);
 int32_t streamProcessDispatchReq(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg* pMsg, bool exec);
 int32_t streamProcessDispatchRsp(SStreamTask* pTask, SStreamDispatchRsp* pRsp);
-int32_t streamProcessRecoverReq(SStreamTask* pTask, SStreamTaskRecoverReq* pReq, SRpcMsg* pMsg);
-int32_t streamProcessRecoverRsp(SStreamTask* pTask, SStreamTaskRecoverRsp* pRsp);

 int32_t streamProcessRetrieveReq(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pMsg);
 int32_t streamProcessRetrieveRsp(SStreamTask* pTask, SStreamRetrieveRsp* pRsp);
@ -468,6 +504,7 @@ typedef struct SStreamMeta {
  TTB*         pTaskDb;
  TTB*         pStateDb;
  SHashObj*    pTasks;
+  SHashObj*    pRecoveringState;
  void*        ahandle;
  TXN          txn;
  FTaskExpand* expandFunc;
@ -484,6 +521,7 @@ SStreamTask* streamMetaGetTask(SStreamMeta* pMeta, int32_t taskId);
 int32_t streamMetaBegin(SStreamMeta* pMeta);
 int32_t streamMetaCommit(SStreamMeta* pMeta);
 int32_t streamMetaRollBack(SStreamMeta* pMeta);
+int32_t streamLoadTasks(SStreamMeta* pMeta);

 #ifdef __cplusplus
 }
--- a/include/libs/stream/tstreamUpdate.h
+++ b/include/libs/stream/tstreamUpdate.h
@ -47,6 +47,8 @@ bool updateInfoIgnore(SUpdateInfo *pInfo, STimeWindow* pWin, uint64_t groupId, u
 void updateInfoDestroy(SUpdateInfo *pInfo);
 void updateInfoAddCloseWindowSBF(SUpdateInfo *pInfo);
 void updateInfoDestoryColseWinSBF(SUpdateInfo *pInfo);
+int32_t updateInfoSerialize(void *buf, int32_t bufLen, const SUpdateInfo *pInfo);
+int32_t updateInfoDeserialize(void *buf, int32_t bufLen, SUpdateInfo *pInfo);

 #ifdef __cplusplus
 }
--- a/include/libs/sync/sync.h
+++ b/include/libs/sync/sync.h
@ -27,6 +27,9 @@ extern "C" {
 extern bool gRaftDetailLog;

 #define SYNC_RESP_TTL_MS       10000000
+#define SYNC_SPEED_UP_HB_TIMER 400
+#define SYNC_SPEED_UP_AFTER_MS (1000 * 20)
+#define SYNC_SLOW_DOWN_RANGE   100

 #define SYNC_MAX_BATCH_SIZE 1
 #define SYNC_INDEX_BEGIN    0
@ -205,6 +208,7 @@ int32_t     syncSetStandby(int64_t rid);
 ESyncState  syncGetMyRole(int64_t rid);
 bool        syncIsReady(int64_t rid);
 const char* syncGetMyRoleStr(int64_t rid);
+bool        syncRestoreFinish(int64_t rid);
 SyncTerm    syncGetMyTerm(int64_t rid);
 SyncGroupId syncGetVgId(int64_t rid);
 void        syncGetEpSet(int64_t rid, SEpSet* pEpSet);
--- a/include/util/tbloomfilter.h
+++ b/include/util/tbloomfilter.h
@ -17,6 +17,7 @@
 #define _TD_UTIL_BLOOMFILTER_H_

 #include "os.h"
+#include "tencode.h"
 #include "thash.h"

 #ifdef __cplusplus
@ -42,6 +43,8 @@ int32_t tBloomFilterNoContain(const SBloomFilter *pBF, const void *keyBuf,
 void tBloomFilterDestroy(SBloomFilter *pBF);
 void tBloomFilterDump(const SBloomFilter *pBF);
 bool tBloomFilterIsFull(const SBloomFilter *pBF);
+int32_t tBloomFilterEncode(const SBloomFilter *pBF, SEncoder* pEncoder);
+SBloomFilter* tBloomFilterDecode(SDecoder* pDecoder);

 #ifdef __cplusplus
 }
--- a/include/util/tdef.h
+++ b/include/util/tdef.h
@ -359,11 +359,11 @@ typedef enum ELogicConditionType {
 #define TSDB_DEFAULT_DB_SCHEMALESS      TSDB_DB_SCHEMALESS_OFF

 #define TSDB_DB_MIN_WAL_RETENTION_PERIOD     -1
-#define TSDB_DEFAULT_DB_WAL_RETENTION_PERIOD 0
+#define TSDB_DEFAULT_DB_WAL_RETENTION_PERIOD (24 * 60 * 60 * 2)
 #define TSDB_DB_MIN_WAL_RETENTION_SIZE       -1
-#define TSDB_DEFAULT_DB_WAL_RETENTION_SIZE   0
+#define TSDB_DEFAULT_DB_WAL_RETENTION_SIZE   -1
 #define TSDB_DB_MIN_WAL_ROLL_PERIOD          0
-#define TSDB_DEFAULT_DB_WAL_ROLL_PERIOD      0
+#define TSDB_DEFAULT_DB_WAL_ROLL_PERIOD      (24 * 60 * 60 * 1)
 #define TSDB_DB_MIN_WAL_SEGMENT_SIZE         0
 #define TSDB_DEFAULT_DB_WAL_SEGMENT_SIZE     0

--- a/include/util/tscalablebf.h
+++ b/include/util/tscalablebf.h
@ -33,7 +33,8 @@ int32_t tScalableBfPut(SScalableBf *pSBf, const void *keyBuf, uint32_t len);
 int32_t tScalableBfNoContain(const SScalableBf *pSBf, const void *keyBuf,
                             uint32_t len);
 void tScalableBfDestroy(SScalableBf *pSBf);
-void tScalableBfDump(const SScalableBf *pSBf);
+int32_t tScalableBfEncode(const SScalableBf *pSBf, SEncoder* pEncoder);
+SScalableBf* tScalableBfDecode(SDecoder* pDecoder);

 #ifdef __cplusplus
 }
--- a/packaging/deb/makedeb.sh
+++ b/packaging/deb/makedeb.sh
@ -60,7 +60,7 @@ cp ${compile_dir}/../packaging/tools/set_core.sh    ${pkg_dir}${install_home_pat
 cp ${compile_dir}/../packaging/tools/taosd-dump-cfg.gdb    ${pkg_dir}${install_home_path}/bin

 cp ${compile_dir}/build/bin/taosd                   ${pkg_dir}${install_home_path}/bin
-#cp ${compile_dir}/build/bin/taosBenchmark           ${pkg_dir}${install_home_path}/bin
+cp ${compile_dir}/build/bin/taosBenchmark           ${pkg_dir}${install_home_path}/bin

 if [ -f "${compile_dir}/build/bin/taosadapter" ]; then
    cp ${compile_dir}/build/bin/taosadapter                    ${pkg_dir}${install_home_path}/bin ||:
--- a/packaging/rpm/tdengine.spec
+++ b/packaging/rpm/tdengine.spec
@ -69,7 +69,7 @@ cp %{_compiledir}/../packaging/tools/set_core.sh    %{buildroot}%{homepath}/bin
 cp %{_compiledir}/../packaging/tools/taosd-dump-cfg.gdb    %{buildroot}%{homepath}/bin
 cp %{_compiledir}/build/bin/taos                    %{buildroot}%{homepath}/bin
 cp %{_compiledir}/build/bin/taosd                   %{buildroot}%{homepath}/bin
-#cp %{_compiledir}/build/bin/taosBenchmark           %{buildroot}%{homepath}/bin
+cp %{_compiledir}/build/bin/taosBenchmark           %{buildroot}%{homepath}/bin

 if [ -f %{_compiledir}/build/bin/taosadapter ]; then
    cp %{_compiledir}/build/bin/taosadapter                    %{buildroot}%{homepath}/bin ||:
--- a/packaging/tools/post.sh
+++ b/packaging/tools/post.sh
@ -132,6 +132,7 @@ function install_bin() {
    [ -x ${bin_dir}/taosd ] && ${csudo}ln -s ${bin_dir}/taosd ${bin_link_dir}/taosd          || :
    [ -x ${bin_dir}/taosadapter ] && ${csudo}ln -s ${bin_dir}/taosadapter ${bin_link_dir}/taosadapter             || :
    [ -x ${bin_dir}/taosBenchmark ] && ${csudo}ln -sf ${bin_dir}/taosBenchmark ${bin_link_dir}/taosdemo || :
+    [ -x ${bin_dir}/taosBenchmark ] && ${csudo}ln -sf ${bin_dir}/taosBenchmark ${bin_link_dir}/taosBenchmark || :
    [ -x ${bin_dir}/TDinsight.sh ] && ${csudo}ln -sf ${bin_dir}/TDinsight.sh ${bin_link_dir}/TDinsight.sh || :
    [ -x ${bin_dir}/taosdump ] && ${csudo}ln -s ${bin_dir}/taosdump ${bin_link_dir}/taosdump || :
    [ -x ${bin_dir}/set_core.sh ] && ${csudo}ln -s ${bin_dir}/set_core.sh ${bin_link_dir}/set_core || :
--- a/source/client/src/TSDBJNIConnector.c
+++ b/source/client/src/TSDBJNIConnector.c
@ -757,7 +757,7 @@ JNIEXPORT jlong JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_prepareStmtImp(J
  int32_t    code = taos_stmt_prepare(pStmt, str, len);
  taosMemoryFreeClear(str);
  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
+    jniError("prepareStmt jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

@ -785,7 +785,7 @@ JNIEXPORT jint JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_setBindTableNameI
  if (code != TSDB_CODE_SUCCESS) {
    (*env)->ReleaseStringUTFChars(env, jname, name);

-    jniError("jobj:%p, conn:%p, code:%s", jobj, tsconn, tstrerror(code));
+    jniError("bindTableName jobj:%p, conn:%p, code:%s", jobj, tsconn, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

@ -860,7 +860,7 @@ JNIEXPORT jint JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_setTableNameTagsI
  (*env)->ReleaseStringUTFChars(env, tableName, name);

  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tsconn, tstrerror(code));
+    jniError("tableNameTags jobj:%p, conn:%p, code:%s", jobj, tsconn, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }
  return JNI_SUCCESS;
@ -926,7 +926,7 @@ JNIEXPORT jlong JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_bindColDataImp(
  taosMemoryFreeClear(b);

  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
+    jniError("bindColData jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

@ -949,7 +949,7 @@ JNIEXPORT jint JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_addBatchImp(JNIEn

  int32_t code = taos_stmt_add_batch(pStmt);
  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
+    jniError("add batch jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

@ -973,7 +973,7 @@ JNIEXPORT jint JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_executeBatchImp(J

  int32_t code = taos_stmt_execute(pStmt);
  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
+    jniError("excute batch jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

@ -997,7 +997,7 @@ JNIEXPORT jint JNICALL Java_com_taosdata_jdbc_TSDBJNIConnector_closeStmt(JNIEnv

  int32_t code = taos_stmt_close(pStmt);
  if (code != TSDB_CODE_SUCCESS) {
-    jniError("jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
+    jniError("close stmt jobj:%p, conn:%p, code:%s", jobj, tscon, tstrerror(code));
    return JNI_TDENGINE_ERROR;
  }

--- a/source/client/src/clientHb.c
+++ b/source/client/src/clientHb.c
@ -327,7 +327,13 @@ int32_t hbBuildQueryDesc(SQueryHbReqBasic *hbBasic, STscObj *pObj) {
  while (pIter != NULL) {
    int64_t     *rid = pIter;
    SRequestObj *pRequest = acquireRequest(*rid);
-    if (NULL == pRequest || pRequest->killed) {
+    if (NULL == pRequest) {
+      pIter = taosHashIterate(pObj->pRequests, pIter);
+      continue;
+    }
+
+    if (pRequest->killed) {
+      releaseRequest(*rid);
      pIter = taosHashIterate(pObj->pRequests, pIter);
      continue;
    }
--- a/source/client/src/clientImpl.c
+++ b/source/client/src/clientImpl.c
@ -283,7 +283,7 @@ void asyncExecLocalCmd(SRequestObj* pRequest, SQuery* pQuery) {

  int32_t code = qExecCommand(pQuery->pRoot, &pRsp);
  if (TSDB_CODE_SUCCESS == code && NULL != pRsp) {
-    code = setQueryResultFromRsp(&pRequest->body.resInfo, pRsp, false, false);
+    code = setQueryResultFromRsp(&pRequest->body.resInfo, pRsp, false, true);
  }

  SReqResultInfo* pResultInfo = &pRequest->body.resInfo;
--- a/source/client/src/clientMain.c
+++ b/source/client/src/clientMain.c
@ -182,6 +182,7 @@ void taos_free_result(TAOS_RES *res) {

  if (TD_RES_QUERY(res)) {
    SRequestObj *pRequest = (SRequestObj *)res;
+    tscDebug("0x%" PRIx64 " taos_free_result start to free query", pRequest->requestId);
    destroyRequest(pRequest);
  } else if (TD_RES_TMQ(res)) {
    SMqRspObj *pRsp = (SMqRspObj *)res;
@ -482,7 +483,7 @@ void taos_stop_query(TAOS_RES *res) {
  int32_t numOfFields = taos_num_fields(pRequest);
  // It is not a query, no need to stop.
  if (numOfFields == 0) {
-    tscDebug("request %" PRIx64 " no need to be killed since not query", pRequest->requestId);
+    tscDebug("request 0x%" PRIx64 " no need to be killed since not query", pRequest->requestId);
    return;
  }

@ -852,7 +853,7 @@ static void fetchCallback(void *pResult, void *param, int32_t code) {
  }

  pRequest->code =
-      setQueryResultFromRsp(pResultInfo, (SRetrieveTableRsp *)pResultInfo->pData, pResultInfo->convertUcs4, false);
+      setQueryResultFromRsp(pResultInfo, (SRetrieveTableRsp *)pResultInfo->pData, pResultInfo->convertUcs4, true);
  if (pRequest->code != TSDB_CODE_SUCCESS) {
    pResultInfo->numOfRows = 0;
    pRequest->code = code;
--- a/source/client/src/clientMsgHandler.c
+++ b/source/client/src/clientMsgHandler.c
@ -389,7 +389,7 @@ int32_t processShowVariablesRsp(void* param, SDataBuf* pMsg, int32_t code) {
      code = buildShowVariablesRsp(rsp.variables, &pRes);
    }
    if (TSDB_CODE_SUCCESS == code) {
-      code = setQueryResultFromRsp(&pRequest->body.resInfo, pRes, false, false);
+      code = setQueryResultFromRsp(&pRequest->body.resInfo, pRes, false, true);
    }

    tFreeSShowVariablesRsp(&rsp);
--- a/source/common/src/systable.c
+++ b/source/common/src/systable.c
@ -24,7 +24,7 @@
 #define SYSTABLE_SCH_COL_NAME_LEN   ((TSDB_COL_NAME_LEN - 1) + VARSTR_HEADER_SIZE)

 static const SSysDbTableSchema dnodesSchema[] = {
-    {.name = "id", .bytes = 4, .type = TSDB_DATA_TYPE_SMALLINT},
+    {.name = "id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
    {.name = "endpoint", .bytes = TSDB_EP_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
    {.name = "vnodes", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT},
    {.name = "support_vnodes", .bytes = 2, .type = TSDB_DATA_TYPE_SMALLINT},
@ -66,7 +66,7 @@ static const SSysDbTableSchema bnodesSchema[] = {
 };

 static const SSysDbTableSchema clusterSchema[] = {
-    {.name = "id", .bytes = 4, .type = TSDB_DATA_TYPE_BIGINT},
+    {.name = "id", .bytes = 4, .type = TSDB_DATA_TYPE_INT},
    {.name = "name", .bytes = TSDB_CLUSTER_ID_LEN + VARSTR_HEADER_SIZE, .type = TSDB_DATA_TYPE_VARCHAR},
    {.name = "create_time", .bytes = 8, .type = TSDB_DATA_TYPE_TIMESTAMP},
 };
--- a/source/common/src/tglobal.c
+++ b/source/common/src/tglobal.c
@ -355,7 +355,11 @@ static int32_t taosAddSystemCfg(SConfig *pCfg) {
 static int32_t taosAddServerCfg(SConfig *pCfg) {
  if (cfgAddDir(pCfg, "dataDir", tsDataDir, 0) != 0) return -1;
  if (cfgAddFloat(pCfg, "minimalDataDirGB", 2.0f, 0.001f, 10000000, 0) != 0) return -1;
+
+  tsNumOfSupportVnodes = tsNumOfCores * 2;
+  tsNumOfSupportVnodes = TMAX(tsNumOfSupportVnodes, 2);
  if (cfgAddInt32(pCfg, "supportVnodes", tsNumOfSupportVnodes, 0, 4096, 0) != 0) return -1;
+
  if (cfgAddInt32(pCfg, "maxShellConns", tsMaxShellConns, 10, 50000000, 0) != 0) return -1;
  if (cfgAddInt32(pCfg, "statusInterval", tsStatusInterval, 1, 30, 0) != 0) return -1;
  if (cfgAddInt32(pCfg, "minSlidingTime", tsMinSlidingTime, 10, 1000000, 0) != 0) return -1;
@ -423,7 +427,7 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
  if (cfgAddInt32(pCfg, "numOfSnodeUniqueThreads", tsNumOfSnodeUniqueThreads, 2, 1024, 0) != 0) return -1;

  tsRpcQueueMemoryAllowed = tsTotalMemoryKB * 1024 * 0.1;
-  tsRpcQueueMemoryAllowed = TRANGE(tsRpcQueueMemoryAllowed, TSDB_MAX_MSG_SIZE * 10L, TSDB_MAX_MSG_SIZE * 10000L);
+  tsRpcQueueMemoryAllowed = TRANGE(tsRpcQueueMemoryAllowed, TSDB_MAX_MSG_SIZE * 10LL, TSDB_MAX_MSG_SIZE * 10000LL);
  if (cfgAddInt64(pCfg, "rpcQueueMemoryAllowed", tsRpcQueueMemoryAllowed, TSDB_MAX_MSG_SIZE * 10L, INT64_MAX, 0) != 0)
    return -1;

--- a/source/dnode/mgmt/node_mgmt/src/dmTransport.c
+++ b/source/dnode/mgmt/node_mgmt/src/dmTransport.c
@ -255,7 +255,8 @@ static inline void dmReleaseHandle(SRpcHandleInfo *pHandle, int8_t type) {
 static bool rpcRfp(int32_t code, tmsg_t msgType) {
  if (code == TSDB_CODE_RPC_REDIRECT || code == TSDB_CODE_RPC_NETWORK_UNAVAIL || code == TSDB_CODE_NODE_NOT_DEPLOYED ||
      code == TSDB_CODE_SYN_NOT_LEADER || code == TSDB_CODE_APP_NOT_READY || code == TSDB_CODE_RPC_BROKEN_LINK) {
-    if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH || msgType == TDMT_SCH_MERGE_FETCH) {
+    if (msgType == TDMT_SCH_QUERY || msgType == TDMT_SCH_MERGE_QUERY || msgType == TDMT_SCH_FETCH ||
+        msgType == TDMT_SCH_MERGE_FETCH) {
      return false;
    }
    return true;
--- a/source/dnode/mnode/impl/inc/mndDb.h
+++ b/source/dnode/mnode/impl/inc/mndDb.h
@ -28,6 +28,7 @@ SDbObj *mndAcquireDb(SMnode *pMnode, const char *db);
 void    mndReleaseDb(SMnode *pMnode, SDbObj *pDb);
 int32_t mndValidateDbInfo(SMnode *pMnode, SDbVgVersion *pDbs, int32_t numOfDbs, void **ppRsp, int32_t *pRspLen);
 int32_t mndExtractDbInfo(SMnode *pMnode, SDbObj *pDb, SUseDbRsp *pRsp, const SUseDbReq *pReq);
+bool    mndIsDbReady(SMnode *pMnode, SDbObj *pDb);

 const char *mndGetDbStr(const char *src);

--- a/source/dnode/mnode/impl/inc/mndDef.h
+++ b/source/dnode/mnode/impl/inc/mndDef.h
@ -162,8 +162,9 @@ typedef struct {
  int64_t     lastExecTime;
  int32_t     lastAction;
  int32_t     lastErrorNo;
-  tmsg_t      lastMsgType;
  SEpSet      lastEpset;
+  tmsg_t      lastMsgType;
+  tmsg_t      originRpcType;
  char        dbname1[TSDB_TABLE_FNAME_LEN];
  char        dbname2[TSDB_TABLE_FNAME_LEN];
  int32_t     startFunc;
--- a/source/dnode/mnode/impl/src/mndDb.c
+++ b/source/dnode/mnode/impl/src/mndDb.c
@ -1495,8 +1495,34 @@ static const char *getCacheModelStr(int8_t cacheModel) {
  return "unknown";
 }

-static void dumpDbInfoData(SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, int32_t rows, int64_t numOfTables,
-                           bool sysDb, ESdbStatus objStatus, bool sysinfo) {
+bool mndIsDbReady(SMnode *pMnode, SDbObj *pDb) {
+  if (pDb->cfg.replications == 1) return true;
+
+  SSdb *pSdb = pMnode->pSdb;
+  void *pIter = NULL;
+  bool  isReady = true;
+  while (1) {
+    SVgObj *pVgroup = NULL;
+    pIter = sdbFetch(pSdb, SDB_VGROUP, pIter, (void **)&pVgroup);
+    if (pIter == NULL) break;
+
+    if (pVgroup->dbUid == pDb->uid && pVgroup->replica > 1) {
+      bool hasLeader = false;
+      for (int32_t i = 0; i < pVgroup->replica; ++i) {
+        if (pVgroup->vnodeGid[i].role == TAOS_SYNC_STATE_LEADER) {
+          hasLeader = true;
+        }
+      }
+      if (!hasLeader) isReady = false;
+    }
+    sdbRelease(pSdb, pVgroup);
+  }
+
+  return isReady;
+}
+
+static void mndDumpDbInfoData(SMnode *pMnode, SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, int32_t rows,
+                              int64_t numOfTables, bool sysDb, ESdbStatus objStatus, bool sysinfo) {
  int32_t cols = 0;
  int32_t bytes = pShow->pMeta->pSchemas[cols].bytes;
  char   *buf = taosMemoryMalloc(bytes);
@ -1509,8 +1535,16 @@ static void dumpDbInfoData(SSDataBlock *pBlock, SDbObj *pDb, SShowObj *pShow, in
  }

  char *statusStr = "ready";
-  if (objStatus == SDB_STATUS_CREATING) statusStr = "creating";
-  if (objStatus == SDB_STATUS_DROPPING) statusStr = "dropping";
+  if (objStatus == SDB_STATUS_CREATING) {
+    statusStr = "creating";
+  } else if (objStatus == SDB_STATUS_DROPPING) {
+    statusStr = "dropping";
+  } else {
+    if (!sysDb && !mndIsDbReady(pMnode, pDb)) {
+      statusStr = "unsynced";
+    }
+  }
+
  char statusVstr[24] = {0};
  STR_WITH_SIZE_TO_VARSTR(statusVstr, statusStr, strlen(statusStr));

@ -1693,7 +1727,7 @@ static int32_t mndRetrieveDbs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
    setInformationSchemaDbCfg(&infoschemaDb);
    size_t numOfTables = 0;
    getInfosDbMeta(NULL, &numOfTables);
-    dumpDbInfoData(pBlock, &infoschemaDb, pShow, numOfRows, numOfTables, true, 0, 1);
+    mndDumpDbInfoData(pMnode, pBlock, &infoschemaDb, pShow, numOfRows, numOfTables, true, 0, 1);

    numOfRows += 1;

@ -1701,7 +1735,7 @@ static int32_t mndRetrieveDbs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
    setPerfSchemaDbCfg(&perfschemaDb);
    numOfTables = 0;
    getPerfDbMeta(NULL, &numOfTables);
-    dumpDbInfoData(pBlock, &perfschemaDb, pShow, numOfRows, numOfTables, true, 0, 1);
+    mndDumpDbInfoData(pMnode, pBlock, &perfschemaDb, pShow, numOfRows, numOfTables, true, 0, 1);

    numOfRows += 1;
    pShow->sysDbRsp = true;
@ -1714,7 +1748,7 @@ static int32_t mndRetrieveDbs(SRpcMsg *pReq, SShowObj *pShow, SSDataBlock *pBloc
    if (mndCheckDbPrivilege(pMnode, pReq->info.conn.user, MND_OPER_READ_OR_WRITE_DB, pDb) == 0) {
      int32_t numOfTables = 0;
      sdbTraverse(pSdb, SDB_VGROUP, mndGetTablesOfDbFp, &numOfTables, NULL, NULL);
-      dumpDbInfoData(pBlock, pDb, pShow, numOfRows, numOfTables, false, objStatus, sysinfo);
+      mndDumpDbInfoData(pMnode, pBlock, pDb, pShow, numOfRows, numOfTables, false, objStatus, sysinfo);
      numOfRows++;
    }

--- a/source/dnode/mnode/impl/src/mndStream.c
+++ b/source/dnode/mnode/impl/src/mndStream.c
@ -504,6 +504,7 @@ static int32_t mndPersistTaskDropReq(STrans *pTrans, SStreamTask *pTask) {
  return 0;
 }

+#if 0
 static int32_t mndPersistTaskRecoverReq(STrans *pTrans, SStreamTask *pTask) {
  SMStreamTaskRecoverReq *pReq = taosMemoryCalloc(1, sizeof(SMStreamTaskRecoverReq));
  if (pReq == NULL) {
@ -540,7 +541,6 @@ static int32_t mndPersistTaskRecoverReq(STrans *pTrans, SStreamTask *pTask) {
  return 0;
 }

-#if 0
 int32_t mndRecoverStreamTasks(SMnode *pMnode, STrans *pTrans, SStreamObj *pStream) {
  if (pStream->isDistributed) {
    int32_t lv = taosArrayGetSize(pStream->tasks);
--- a/source/dnode/mnode/impl/src/mndTrans.c
+++ b/source/dnode/mnode/impl/src/mndTrans.c
@ -124,8 +124,7 @@ static SSdbRaw *mndTransActionEncode(STrans *pTrans) {
  SDB_SET_INT8(pRaw, dataPos, pTrans->exec, _OVER)
  SDB_SET_INT8(pRaw, dataPos, pTrans->oper, _OVER)
  SDB_SET_INT8(pRaw, dataPos, 0, _OVER)
-  SDB_SET_INT8(pRaw, dataPos, 0, _OVER)
-  SDB_SET_INT8(pRaw, dataPos, 0, _OVER)
+  SDB_SET_INT16(pRaw, dataPos, pTrans->originRpcType, _OVER)
  SDB_SET_INT64(pRaw, dataPos, pTrans->createdTime, _OVER)
  SDB_SET_BINARY(pRaw, dataPos, pTrans->dbname1, TSDB_TABLE_FNAME_LEN, _OVER)
  SDB_SET_BINARY(pRaw, dataPos, pTrans->dbname2, TSDB_TABLE_FNAME_LEN, _OVER)
@ -282,13 +281,12 @@ static SSdbRow *mndTransActionDecode(SSdbRaw *pRaw) {
  SDB_GET_INT8(pRaw, dataPos, &exec, _OVER)
  SDB_GET_INT8(pRaw, dataPos, &oper, _OVER)
  SDB_GET_INT8(pRaw, dataPos, &reserved, _OVER)
-  SDB_GET_INT8(pRaw, dataPos, &reserved, _OVER)
-  SDB_GET_INT8(pRaw, dataPos, &reserved, _OVER)
  pTrans->stage = stage;
  pTrans->policy = policy;
  pTrans->conflict = conflict;
  pTrans->exec = exec;
  pTrans->oper = oper;
+  SDB_GET_INT16(pRaw, dataPos, &pTrans->originRpcType, _OVER)
  SDB_GET_INT64(pRaw, dataPos, &pTrans->createdTime, _OVER)
  SDB_GET_BINARY(pRaw, dataPos, pTrans->dbname1, TSDB_TABLE_FNAME_LEN, _OVER)
  SDB_GET_BINARY(pRaw, dataPos, pTrans->dbname2, TSDB_TABLE_FNAME_LEN, _OVER)
@ -611,6 +609,7 @@ STrans *mndTransCreate(SMnode *pMnode, ETrnPolicy policy, ETrnConflct conflict,

  if (pReq != NULL) {
    taosArrayPush(pTrans->pRpcArray, &pReq->info);
+    pTrans->originRpcType = pReq->msgType;
  }
  mTrace("trans:%d, local object is created, data:%p", pTrans->id, pTrans);
  return pTrans;
@ -910,6 +909,23 @@ static void mndTransSendRpcRsp(SMnode *pMnode, STrans *pTrans) {
        }
      }

+      if (pTrans->originRpcType == TDMT_MND_CREATE_DB) {
+        mDebug("trans:%d, origin msgtype:%s", pTrans->id, TMSG_INFO(pTrans->originRpcType));
+        SDbObj *pDb = mndAcquireDb(pMnode, pTrans->dbname1);
+        if (pDb != NULL) {
+          for (int32_t j = 0; j < 12; j++) {
+            bool ready = mndIsDbReady(pMnode, pDb);
+            if (!ready) {
+              mDebug("trans:%d, db:%s not ready yet, wait %d times", pTrans->id, pTrans->dbname1, j);
+              taosMsleep(1000);
+            } else {
+              break;
+            }
+          }
+        }
+        mndReleaseDb(pMnode, pDb);
+      }
+
      tmsgSendRsp(&rspMsg);
    }
  }
--- a/source/dnode/vnode/src/inc/vnodeInt.h
+++ b/source/dnode/vnode/src/inc/vnodeInt.h
@ -97,7 +97,7 @@ typedef struct STbUidStore STbUidStore;

 int             metaOpen(SVnode* pVnode, SMeta** ppMeta);
 int             metaClose(SMeta* pMeta);
-int             metaBegin(SMeta* pMeta);
+int             metaBegin(SMeta* pMeta, int8_t fromSys);
 int             metaCommit(SMeta* pMeta);
 int             metaCreateSTable(SMeta* pMeta, int64_t version, SVCreateStbReq* pReq);
 int             metaAlterSTable(SMeta* pMeta, int64_t version, SVCreateStbReq* pReq);
@ -187,7 +187,7 @@ int32_t smaAsyncPreCommit(SSma* pSma);
 int32_t smaAsyncCommit(SSma* pSma);
 int32_t smaAsyncPostCommit(SSma* pSma);
 int32_t smaDoRetention(SSma* pSma, int64_t now);
-int32_t smaProcessFetch(SSma *pSma, void* pMsg);
+int32_t smaProcessFetch(SSma* pSma, void* pMsg);

 int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg);
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg);
--- a/source/dnode/vnode/src/meta/metaCommit.c
+++ b/source/dnode/vnode/src/meta/metaCommit.c
@ -19,9 +19,12 @@ static FORCE_INLINE void *metaMalloc(void *pPool, size_t size) { return vnodeBuf
 static FORCE_INLINE void  metaFree(void *pPool, void *p) { vnodeBufPoolFree((SVBufPool *)pPool, p); }

 // begin a meta txn
-int metaBegin(SMeta *pMeta) {
+int metaBegin(SMeta *pMeta, int8_t fromSys) {
+  if (fromSys) {
+    tdbTxnOpen(&pMeta->txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED);
+  } else {
    tdbTxnOpen(&pMeta->txn, 0, metaMalloc, metaFree, pMeta->pVnode->inUse, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED);
-
+  }
  if (tdbBegin(pMeta->pEnv, &pMeta->txn) < 0) {
    return -1;
  }
--- a/source/dnode/vnode/src/meta/metaSnapshot.c
+++ b/source/dnode/vnode/src/meta/metaSnapshot.c
@ -145,7 +145,7 @@ int32_t metaSnapWriterOpen(SMeta* pMeta, int64_t sver, int64_t ever, SMetaSnapWr
  pWriter->sver = sver;
  pWriter->ever = ever;

-  metaBegin(pMeta);
+  metaBegin(pMeta, 1);

  *ppWriter = pWriter;
  return code;
--- a/source/dnode/vnode/src/tq/tq.c
+++ b/source/dnode/vnode/src/tq/tq.c
@ -777,6 +777,7 @@ int32_t tqProcessTaskDispatchReq(STQ* pTq, SRpcMsg* pMsg, bool exec) {
  }
 }

+#if 0
 int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
  SStreamTaskRecoverReq* pReq = pMsg->pCont;
  int32_t                taskId = pReq->taskId;
@ -789,18 +790,6 @@ int32_t tqProcessTaskRecoverReq(STQ* pTq, SRpcMsg* pMsg) {
  }
 }

-int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
-  SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
-  int32_t             taskId = pRsp->taskId;
-  SStreamTask*        pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
-  if (pTask) {
-    streamProcessDispatchRsp(pTask, pRsp);
-    return 0;
-  } else {
-    return -1;
-  }
-}
-
 int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
  SStreamTaskRecoverRsp* pRsp = pMsg->pCont;
  int32_t                taskId = pRsp->rspTaskId;
@ -813,6 +802,19 @@ int32_t tqProcessTaskRecoverRsp(STQ* pTq, SRpcMsg* pMsg) {
    return -1;
  }
 }
+#endif
+
+int32_t tqProcessTaskDispatchRsp(STQ* pTq, SRpcMsg* pMsg) {
+  SStreamDispatchRsp* pRsp = POINTER_SHIFT(pMsg->pCont, sizeof(SMsgHead));
+  int32_t             taskId = pRsp->taskId;
+  SStreamTask*        pTask = streamMetaGetTask(pTq->pStreamMeta, taskId);
+  if (pTask) {
+    streamProcessDispatchRsp(pTask, pRsp);
+    return 0;
+  } else {
+    return -1;
+  }
+}

 int32_t tqProcessTaskDropReq(STQ* pTq, char* msg, int32_t msgLen) {
  SVDropStreamTaskReq* pReq = (SVDropStreamTaskReq*)msg;
@ -859,8 +861,10 @@ void vnodeEnqueueStreamMsg(SVnode* pVnode, SRpcMsg* pMsg) {
  tDecoderInit(&decoder, msgBody, msgLen);
  if (tDecodeStreamDispatchReq(&decoder, &req) < 0) {
    code = TSDB_CODE_MSG_DECODE_ERROR;
+    tDecoderClear(&decoder);
    goto FAIL;
  }
+  tDecoderClear(&decoder);

  int32_t taskId = req.taskId;

@ -871,6 +875,8 @@ void vnodeEnqueueStreamMsg(SVnode* pVnode, SRpcMsg* pMsg) {
        .code = 0,
    };
    streamProcessDispatchReq(pTask, &req, &rsp, false);
+    rpcFreeCont(pMsg->pCont);
+    taosFreeQitem(pMsg);
    return;
  }

@ -881,4 +887,6 @@ FAIL:
      .info = pMsg->info,
  };
  tmsgSendRsp(&rsp);
+  rpcFreeCont(pMsg->pCont);
+  taosFreeQitem(pMsg);
 }
--- a/source/dnode/vnode/src/vnd/vnodeCommit.c
+++ b/source/dnode/vnode/src/vnd/vnodeCommit.c
@ -42,7 +42,7 @@ int vnodeBegin(SVnode *pVnode) {

  pVnode->state.commitID++;
  // begin meta
-  if (metaBegin(pVnode->pMeta) < 0) {
+  if (metaBegin(pVnode->pMeta, 0) < 0) {
    vError("vgId:%d, failed to begin meta since %s", TD_VID(pVnode), tstrerror(terrno));
    return -1;
  }
--- a/source/dnode/vnode/src/vnd/vnodeSvr.c
+++ b/source/dnode/vnode/src/vnd/vnodeSvr.c
@ -334,14 +334,14 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
    case TDMT_STREAM_TASK_DISPATCH:
      //      return tqProcessTaskDispatchReq(pVnode->pTq, pMsg, pInfo->workerId != 0);
      return tqProcessTaskDispatchReq(pVnode->pTq, pMsg, true);
-    case TDMT_STREAM_TASK_RECOVER:
-      return tqProcessTaskRecoverReq(pVnode->pTq, pMsg);
+      /*case TDMT_STREAM_TASK_RECOVER:*/
+      /*return tqProcessTaskRecoverReq(pVnode->pTq, pMsg);*/
    case TDMT_STREAM_RETRIEVE:
      return tqProcessTaskRetrieveReq(pVnode->pTq, pMsg);
    case TDMT_STREAM_TASK_DISPATCH_RSP:
      return tqProcessTaskDispatchRsp(pVnode->pTq, pMsg);
-    case TDMT_STREAM_TASK_RECOVER_RSP:
-      return tqProcessTaskRecoverRsp(pVnode->pTq, pMsg);
+      /*case TDMT_STREAM_TASK_RECOVER_RSP:*/
+      /*return tqProcessTaskRecoverRsp(pVnode->pTq, pMsg);*/
    case TDMT_STREAM_RETRIEVE_RSP:
      return tqProcessTaskRetrieveRsp(pVnode->pTq, pMsg);
    default:
--- a/source/dnode/vnode/src/vnd/vnodeSync.c
+++ b/source/dnode/vnode/src/vnd/vnodeSync.c
@ -141,6 +141,10 @@ static void inline vnodeHandleWriteMsg(SVnode *pVnode, SRpcMsg *pMsg) {
  }
  if (rsp.info.handle != NULL) {
    tmsgSendRsp(&rsp);
+  } else {
+    if (rsp.pCont) {
+      rpcFreeCont(rsp.pCont);
+    }
  }
 }

@ -299,6 +303,10 @@ void vnodeApplyWriteMsg(SQueueInfo *pInfo, STaosQall *qall, int32_t numOfMsgs) {
    vnodePostBlockMsg(pVnode, pMsg);
    if (rsp.info.handle != NULL) {
      tmsgSendRsp(&rsp);
+    } else {
+      if (rsp.pCont) {
+        rpcFreeCont(rsp.pCont);
+      }
    }

    vGTrace("vgId:%d, msg:%p is freed, code:0x%x index:%" PRId64, vgId, pMsg, rsp.code, pMsg->info.conn.applyIndex);
@ -722,7 +730,8 @@ void vnodeSyncClose(SVnode *pVnode) { syncStop(pVnode->sync); }

 bool vnodeIsLeader(SVnode *pVnode) {
  if (!syncIsReady(pVnode->sync)) {
-    vDebug("vgId:%d, vnode not ready", pVnode->config.vgId);
+    vDebug("vgId:%d, vnode not ready, state:%s, restore:%d", pVnode->config.vgId, syncGetMyRoleStr(pVnode->sync),
+           syncRestoreFinish(pVnode->sync));
    return false;
  }

--- a/source/libs/catalog/inc/catalogInt.h
+++ b/source/libs/catalog/inc/catalogInt.h
@ -279,6 +279,7 @@ typedef struct SCtgMsgCtx {
  void* lastOut;
  void* out;
  char* target;
+  SHashObj* pBatchs;  
 } SCtgMsgCtx;


@ -315,7 +316,6 @@ typedef struct SCtgTask {
  SRWLatch        lock;
  SArray*         pParents;
  SCtgSubRes      subRes;
-  SHashObj*       pBatchs;
 } SCtgTask;

 typedef struct SCtgTaskReq {
--- a/source/libs/catalog/src/ctgAsync.c
+++ b/source/libs/catalog/src/ctgAsync.c
@ -855,6 +855,7 @@ int32_t ctgCallSubCb(SCtgTask *pTask) {
  
  int32_t parentNum = taosArrayGetSize(pTask->pParents);
  for (int32_t i = 0; i < parentNum; ++i) {
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
    SCtgTask* pParent = taosArrayGetP(pTask->pParents, i);
    
    pParent->subRes.code = pTask->code;
@ -865,7 +866,9 @@ int32_t ctgCallSubCb(SCtgTask *pTask) {
      }
    }

-    pParent->pBatchs = pTask->pBatchs;    
+    SCtgMsgCtx *pParMsgCtx = CTG_GET_TASK_MSGCTX(pParent, -1);
+
+    pParMsgCtx->pBatchs = pMsgCtx->pBatchs;    
    CTG_ERR_JRET(pParent->subRes.fp(pParent));
  }
  
@ -1625,6 +1628,11 @@ _return:
 int32_t ctgLaunchGetTbMetaTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetTbMetaFromCache(pCtg, pConn, (SCtgTbMetaCtx*)pTask->taskCtx, (STableMeta**)&pTask->res));
  if (pTask->res) {
@ -1645,6 +1653,7 @@ int32_t ctgLaunchGetTbMetasTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgTbMetasCtx* pCtx = (SCtgTbMetasCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;

  int32_t dbNum = taosArrayGetSize(pCtx->pNames);
  int32_t fetchIdx = 0;
@ -1670,6 +1679,10 @@ int32_t ctgLaunchGetTbMetasTask(SCtgTask *pTask) {
  for (int32_t i = 0; i < pCtx->fetchNum; ++i) {
    SCtgFetch* pFetch = taosArrayGet(pCtx->pFetchs, i);
    SName* pName = ctgGetFetchName(pCtx->pNames, pFetch);
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, i);
+    if (NULL == pMsgCtx->pBatchs) {
+      pMsgCtx->pBatchs = pJob->pBatchs;
+    }
    
    SCtgTaskReq tReq;
    tReq.pTask = pTask;
@ -1686,6 +1699,11 @@ int32_t ctgLaunchGetDbVgTask(SCtgTask *pTask) {
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgDBCache *dbCache = NULL;
  SCtgDbVgCtx* pCtx = (SCtgDbVgCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }
  
  CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, pCtx->dbFName, &dbCache));
  if (NULL != dbCache) {
@ -1722,6 +1740,11 @@ int32_t ctgLaunchGetTbHashTask(SCtgTask *pTask) {
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgDBCache *dbCache = NULL;
  SCtgTbHashCtx* pCtx = (SCtgTbHashCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }
  
  CTG_ERR_RET(ctgAcquireVgInfoFromCache(pCtg, pCtx->dbFName, &dbCache));
  if (NULL != dbCache) {
@ -1761,6 +1784,7 @@ int32_t ctgLaunchGetTbHashsTask(SCtgTask *pTask) {
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgTbHashsCtx* pCtx = (SCtgTbHashsCtx*)pTask->taskCtx;
  SCtgDBCache *dbCache = NULL;
+  SCtgJob* pJob = pTask->pJob;  
  int32_t dbNum = taosArrayGetSize(pCtx->pNames);
  int32_t fetchIdx = 0;
  int32_t baseResIdx = 0;
@ -1803,6 +1827,10 @@ int32_t ctgLaunchGetTbHashsTask(SCtgTask *pTask) {
  for (int32_t i = 0; i < pCtx->fetchNum; ++i) {
    SCtgFetch* pFetch = taosArrayGet(pCtx->pFetchs, i);
    STablesReq* pReq = taosArrayGet(pCtx->pNames, pFetch->dbIdx);
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, i);
+    if (NULL == pMsgCtx->pBatchs) {
+      pMsgCtx->pBatchs = pJob->pBatchs;
+    }
    
    SBuildUseDBInput input = {0};
    strcpy(input.db, pReq->dbFName);
@ -1831,6 +1859,11 @@ int32_t ctgLaunchGetTbIndexTask(SCtgTask *pTask) {
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgTbIndexCtx* pCtx = (SCtgTbIndexCtx*)pTask->taskCtx;
  SArray* pRes = NULL;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgReadTbIndexFromCache(pCtg, pCtx->pName, &pRes));
  if (pRes) {
@ -1852,6 +1885,11 @@ int32_t ctgLaunchGetTbCfgTask(SCtgTask *pTask) {
  SArray* pRes = NULL;
  char dbFName[TSDB_DB_FNAME_LEN];
  tNameGetFullDbName(pCtx->pName, dbFName);
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  if (pCtx->tbType <= 0) {
    CTG_ERR_JRET(ctgReadTbTypeFromCache(pCtg, dbFName, pCtx->pName->tname, &pCtx->tbType));
@ -1890,6 +1928,11 @@ _return:
 int32_t ctgLaunchGetQnodeTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetQnodeListFromMnode(pCtg, pConn, NULL, pTask));
  return TSDB_CODE_SUCCESS;
@ -1898,6 +1941,11 @@ int32_t ctgLaunchGetQnodeTask(SCtgTask *pTask) {
 int32_t ctgLaunchGetDnodeTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetDnodeListFromMnode(pCtg, pConn, NULL, pTask));
  return TSDB_CODE_SUCCESS;
@ -1908,6 +1956,11 @@ int32_t ctgLaunchGetDbCfgTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgDbCfgCtx* pCtx = (SCtgDbCfgCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetDBCfgFromMnode(pCtg, pConn, pCtx->dbFName, NULL, pTask));

@ -1919,6 +1972,11 @@ int32_t ctgLaunchGetDbInfoTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SCtgDBCache *dbCache = NULL;
  SCtgDbInfoCtx* pCtx = (SCtgDbInfoCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  pTask->res = taosMemoryCalloc(1, sizeof(SDbInfo));
  if (NULL == pTask->res) {
@ -1953,6 +2011,11 @@ int32_t ctgLaunchGetIndexTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgIndexCtx* pCtx = (SCtgIndexCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetIndexInfoFromMnode(pCtg, pConn, pCtx->indexFName, NULL, pTask));

@ -1963,6 +2026,11 @@ int32_t ctgLaunchGetUdfTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
  SCtgUdfCtx* pCtx = (SCtgUdfCtx*)pTask->taskCtx;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetUdfInfoFromMnode(pCtg, pConn, pCtx->udfName, NULL, pTask));

@ -1975,6 +2043,11 @@ int32_t ctgLaunchGetUserTask(SCtgTask *pTask) {
  SCtgUserCtx* pCtx = (SCtgUserCtx*)pTask->taskCtx;
  bool inCache = false;
  bool pass = false;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }
  
  CTG_ERR_RET(ctgChkAuthFromCache(pCtg, pCtx->user.user, pCtx->user.dbFName, pCtx->user.type, &inCache, &pass));
  if (inCache) {
@ -1996,6 +2069,11 @@ int32_t ctgLaunchGetUserTask(SCtgTask *pTask) {
 int32_t ctgLaunchGetSvrVerTask(SCtgTask *pTask) {
  SCatalog* pCtg = pTask->pJob->pCtg; 
  SRequestConnInfo* pConn = &pTask->pJob->conn;
+  SCtgJob* pJob = pTask->pJob;
+  SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+  if (NULL == pMsgCtx->pBatchs) {
+    pMsgCtx->pBatchs = pJob->pBatchs;
+  }

  CTG_ERR_RET(ctgGetSvrVerFromMnode(pCtg, pConn, NULL, pTask));

@ -2129,7 +2207,10 @@ int32_t ctgSetSubTaskCb(SCtgTask *pSub, SCtgTask *pTask) {
  if (CTG_TASK_DONE == pSub->status) {
    pTask->subRes.code = pSub->code;
    CTG_ERR_JRET((*gCtgAsyncFps[pTask->type].cloneFp)(pSub, &pTask->subRes.res));
-    pTask->pBatchs = pSub->pBatchs;
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+    SCtgMsgCtx *pSubMsgCtx = CTG_GET_TASK_MSGCTX(pSub, -1);
+    pMsgCtx->pBatchs = pSubMsgCtx->pBatchs;
+    
    CTG_ERR_JRET(pTask->subRes.fp(pTask));
  } else {
    if (NULL == pSub->pParents) {
@ -2167,7 +2248,10 @@ int32_t ctgLaunchSubTask(SCtgTask *pTask, CTG_TASK_TYPE type, ctgSubTaskCbFp fp,
  CTG_ERR_RET(ctgSetSubTaskCb(pSub, pTask));

  if (newTask) {
-    pSub->pBatchs = pTask->pBatchs;
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+    SCtgMsgCtx *pSubMsgCtx = CTG_GET_TASK_MSGCTX(pSub, -1);
+    pSubMsgCtx->pBatchs = pMsgCtx->pBatchs;
+
    CTG_ERR_RET((*gCtgAsyncFps[pSub->type].launchFp)(pSub));
    pSub->status = CTG_TASK_LAUNCHED;
  }
@ -2180,7 +2264,6 @@ int32_t ctgLaunchJob(SCtgJob *pJob) {
  
  for (int32_t i = 0; i < taskNum; ++i) {
    SCtgTask *pTask = taosArrayGet(pJob->pTasks, i);
-    pTask->pBatchs = pJob->pBatchs;

    qDebug("QID:0x%" PRIx64 " ctg launch [%dth] task", pJob->queryId, pTask->taskId);
    CTG_ERR_RET((*gCtgAsyncFps[pTask->type].launchFp)(pTask));
--- a/source/libs/catalog/src/ctgDbg.c
+++ b/source/libs/catalog/src/ctgDbg.c
@ -19,7 +19,7 @@
 #include "catalogInt.h"

 extern SCatalogMgmt gCtgMgmt;
-SCtgDebug gCTGDebug = {.lockEnable = true};
+SCtgDebug gCTGDebug = {0};

 void ctgdUserCallback(SMetaData* pResult, void* param, int32_t code) {
  ASSERT(*(int32_t*)param == 1);
--- a/source/libs/catalog/src/ctgRemote.c
+++ b/source/libs/catalog/src/ctgRemote.c
@ -69,13 +69,13 @@ int32_t ctgHandleBatchRsp(SCtgJob* pJob, SCtgTaskCallbackParam* cbParam, SDataBu
      taskMsg.len = 0;
    }
    
-    pTask->pBatchs = pBatchs;
-    
    SCtgTaskReq tReq;
    tReq.pTask = pTask;
    tReq.msgIdx = rsp.msgIdx;    
+    SCtgMsgCtx* pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, tReq.msgIdx);
+    pMsgCtx->pBatchs = pBatchs;

-    ctgDebug("QID:0x%" PRIx64 " ctg task %d idx %d start to handle rsp %s", pJob->queryId, pTask->taskId, rsp.msgIdx, TMSG_INFO(taskMsg.msgType + 1));
+    ctgDebug("QID:0x%" PRIx64 " ctg task %d idx %d start to handle rsp %s, pBatchs: %p", pJob->queryId, pTask->taskId, rsp.msgIdx, TMSG_INFO(taskMsg.msgType + 1), pBatchs);

    (*gCtgAsyncFps[pTask->type].handleRspFp)(&tReq, rsp.reqType, &taskMsg, (rsp.rspCode ? rsp.rspCode : rspCode));
  }
@ -343,7 +343,9 @@ int32_t ctgHandleMsgCallback(void* param, SDataBuf* pMsg, int32_t rspCode) {
      ctgError("taosHashInit %d batch failed", CTG_DEFAULT_BATCH_NUM);
      CTG_ERR_JRET(TSDB_CODE_OUT_OF_MEMORY);
    }
-    pTask->pBatchs = pBatchs;
+
+    SCtgMsgCtx *pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, -1);
+    pMsgCtx->pBatchs = pBatchs;
 #endif

    SCtgTaskReq tReq;
@ -444,7 +446,8 @@ int32_t ctgAddBatch(SCatalog* pCtg, int32_t vgId, SRequestConnInfo* pConn, SCtgT
                    uint32_t msgSize) {
  int32_t    code = 0;
  SCtgTask*  pTask = tReq->pTask;
-  SHashObj*  pBatchs = pTask->pBatchs;
+  SCtgMsgCtx* pMsgCtx = CTG_GET_TASK_MSGCTX(pTask, tReq->msgIdx);
+  SHashObj*  pBatchs = pMsgCtx->pBatchs;
  SCtgJob*   pJob = pTask->pJob;
  SCtgBatch* pBatch = taosHashGet(pBatchs, &vgId, sizeof(vgId));
  SCtgBatch newBatch = {0};
--- a/source/libs/executor/inc/executorimpl.h
+++ b/source/libs/executor/inc/executorimpl.h
@ -738,6 +738,7 @@ typedef struct STimeSliceOperatorInfo {
  int64_t                 current;
  SArray*                 pPrevRow;      // SArray<SGroupValue>
  SArray*                 pNextRow;      // SArray<SGroupValue>
+  SArray*                 pLinearInfo;   // SArray<SFillLinearInfo>
  bool                    isPrevRowSet;
  bool                    isNextRowSet;
  int32_t                 fillType;      // fill type
--- a/source/libs/executor/inc/tfill.h
+++ b/source/libs/executor/inc/tfill.h
@ -33,6 +33,15 @@ typedef struct SFillColInfo {
  SVariant   fillVal;
 } SFillColInfo;

+typedef struct SFillLinearInfo {
+  SPoint  start;
+  SPoint  end;
+  bool    hasNull;
+  bool    fillLastPoint;
+  int16_t type;
+  int32_t bytes;
+} SFillLinearInfo;
+
 typedef struct {
  SSchema col;
  char*   tagVal;
--- a/source/libs/executor/src/scanoperator.c
+++ b/source/libs/executor/src/scanoperator.c
@ -2204,10 +2204,10 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
  // build message and send to mnode to fetch the content of system tables.
  SExecTaskInfo*     pTaskInfo = pOperator->pTaskInfo;
  SSysTableScanInfo* pInfo = pOperator->info;
+  char dbName[TSDB_DB_NAME_LEN] = {0};

  const char* name = tNameGetTableName(&pInfo->name);
  if (pInfo->showRewrite) {
-    char dbName[TSDB_DB_NAME_LEN] = {0};
    getDBNameFromCondition(pInfo->pCondition, dbName);
    sprintf(pInfo->req.db, "%d.%s", pInfo->accountId, dbName);
  }
@ -2217,7 +2217,7 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
  } else if (strncasecmp(name, TSDB_INS_TABLE_TAGS, TSDB_TABLE_FNAME_LEN) == 0) {
    return sysTableScanUserTags(pOperator);
  } else if (strncasecmp(name, TSDB_INS_TABLE_STABLES, TSDB_TABLE_FNAME_LEN) == 0 &&
-             IS_SYS_DBNAME(pInfo->req.db)) {
+             pInfo->showRewrite && IS_SYS_DBNAME(dbName)) {
    return sysTableScanUserSTables(pOperator);
  } else {  // load the meta from mnode of the given epset
    if (pOperator->status == OP_EXEC_DONE) {
--- a/source/libs/executor/src/timewindowoperator.c
+++ b/source/libs/executor/src/timewindowoperator.c
@ -2087,6 +2087,34 @@ static void doKeepNextRows(STimeSliceOperatorInfo* pSliceInfo, const SSDataBlock
  pSliceInfo->isNextRowSet = true;
 }

+static void doKeepLinearInfo(STimeSliceOperatorInfo* pSliceInfo, const SSDataBlock* pBlock, int32_t rowIndex) {
+  int32_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    SColumnInfoData* pColInfoData = taosArrayGet(pBlock->pDataBlock, i);
+    SColumnInfoData* pTsCol = taosArrayGet(pBlock->pDataBlock, pSliceInfo->tsCol.slotId);
+    SFillLinearInfo* pLinearInfo = taosArrayGet(pSliceInfo->pLinearInfo, i);
+
+    // null data should not be kept since it can not be used to perform interpolation
+    if (!colDataIsNull_s(pColInfoData, i)) {
+      int64_t startKey = *(int64_t*)colDataGetData(pTsCol, rowIndex);
+      int64_t endKey   = *(int64_t*)colDataGetData(pTsCol, rowIndex + 1);
+      pLinearInfo->start.key = startKey;
+      pLinearInfo->end.key   = endKey;
+
+      char* val;
+      val = colDataGetData(pColInfoData, rowIndex);
+      memcpy(pLinearInfo->start.val, val, pLinearInfo->bytes);
+      val = colDataGetData(pColInfoData, rowIndex + 1);
+      memcpy(pLinearInfo->end.val, val, pLinearInfo->bytes);
+
+      pLinearInfo->hasNull = false;
+    } else {
+      pLinearInfo->hasNull = true;
+    }
+  }
+
+}
+
 static void genInterpolationResult(STimeSliceOperatorInfo* pSliceInfo, SExprSupp* pExprSup, SSDataBlock* pBlock,
                                   SSDataBlock* pResBlock) {
  int32_t rows = pResBlock->info.rows;
@ -2115,52 +2143,41 @@ static void genInterpolationResult(STimeSliceOperatorInfo* pSliceInfo, SExprSupp
        if (pDst->info.type == TSDB_DATA_TYPE_FLOAT) {
          float v = 0;
          GET_TYPED_DATA(v, float, pVar->nType, &pVar->i);
-          colDataAppend(pDst, rows, (char*)&v, false);
+          colDataAppend(pDst, rows, (char *)&v, false);
        } else if (pDst->info.type == TSDB_DATA_TYPE_DOUBLE) {
          double v = 0;
          GET_TYPED_DATA(v, double, pVar->nType, &pVar->i);
-          colDataAppend(pDst, rows, (char*)&v, false);
+          colDataAppend(pDst, rows, (char *)&v, false);
        } else if (IS_SIGNED_NUMERIC_TYPE(pDst->info.type)) {
          int64_t v = 0;
          GET_TYPED_DATA(v, int64_t, pVar->nType, &pVar->i);
-          colDataAppend(pDst, rows, (char*)&v, false);
+          colDataAppend(pDst, rows, (char *)&v, false);
        }
        pResBlock->info.rows += 1;
        break;
      }

      case TSDB_FILL_LINEAR: {
-#if 0
-        if (pCtx->start.key == INT64_MIN || pCtx->start.key > pCtx->startTs
-                    || pCtx->end.key == INT64_MIN || pCtx->end.key < pCtx->startTs) {
-//                  goto interp_exit;
+        SFillLinearInfo* pLinearInfo = taosArrayGet(pSliceInfo->pLinearInfo, srcSlot);
+
+        SPoint start   = pLinearInfo->start;
+        SPoint end     = pLinearInfo->end;
+        SPoint current = {.key = pSliceInfo->current};
+        current.val = taosMemoryCalloc(pLinearInfo->bytes, 1);
+
+        // before interp range, do not fill
+        if (start.key == INT64_MIN || end.key == INT64_MAX) {
+          break;
        }

-              double v1 = -1, v2 = -1;
-              GET_TYPED_DATA(v1, double, pCtx->inputType, &pCtx->start.val);
-              GET_TYPED_DATA(v2, double, pCtx->inputType, &pCtx->end.val);
-
-              SPoint point1 = {.key = ts, .val = &v1};
-              SPoint point2 = {.key = nextTs, .val = &v2};
-              SPoint point  = {.key = pCtx->startTs, .val = pCtx->pOutput};
-
-              int32_t srcType = pCtx->inputType;
-              if (isNull((char *)&pCtx->start.val, srcType) || isNull((char *)&pCtx->end.val, srcType)) {
-                setNull(pCtx->pOutput, srcType, pCtx->inputBytes);
+        if (pLinearInfo->hasNull) {
+          colDataAppendNULL(pDst, rows);
        } else {
-                bool exceedMax = false, exceedMin = false;
-                taosGetLinearInterpolationVal(&point, pCtx->outputType, &point1, &point2, TSDB_DATA_TYPE_DOUBLE, &exceedMax, &exceedMin);
-                if (exceedMax || exceedMin) {
-                  __compar_fn_t func = getComparFunc((int32_t)pCtx->inputType, 0);
-                  if (func(&pCtx->start.val, &pCtx->end.val) <= 0) {
-                    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->start.val : &pCtx->end.val);
-                  } else {
-                    COPY_TYPED_DATA(pCtx->pOutput, pCtx->inputType, exceedMax ? &pCtx->end.val : &pCtx->start.val);
+          taosGetLinearInterpolationVal(&current, pLinearInfo->type, &start, &end, pLinearInfo->type);
+          colDataAppend(pDst, rows, (char *)current.val, false);
        }
-                }
-              }
-#endif
-        // TODO: pResBlock->info.rows += 1;
+
+        pResBlock->info.rows += 1;
        break;
      }
      case TSDB_FILL_PREV: {
@ -2246,6 +2263,55 @@ static int32_t initNextRowsKeeper(STimeSliceOperatorInfo* pInfo, SSDataBlock* pB
  return TSDB_CODE_SUCCESS;
 }

+static int32_t initFillLinearInfo(STimeSliceOperatorInfo* pInfo, SSDataBlock* pBlock) {
+  if (pInfo->pLinearInfo != NULL) {
+    return TSDB_CODE_SUCCESS;
+  }
+
+  pInfo->pLinearInfo = taosArrayInit(4, sizeof(SFillLinearInfo));
+  if (pInfo->pNextRow == NULL) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+
+  int32_t numOfCols = taosArrayGetSize(pBlock->pDataBlock);
+  for (int32_t i = 0; i < numOfCols; ++i) {
+    SColumnInfoData* pColInfo = taosArrayGet(pBlock->pDataBlock, i);
+
+    SFillLinearInfo linearInfo = {0};
+    linearInfo.start.key = INT64_MIN;
+    linearInfo.end.key   = INT64_MAX;
+    linearInfo.start.val = taosMemoryCalloc(1, pColInfo->info.bytes);
+    linearInfo.end.val   = taosMemoryCalloc(1, pColInfo->info.bytes);
+    linearInfo.hasNull   = false;
+    linearInfo.fillLastPoint = false;
+    linearInfo.type  = pColInfo->info.type;
+    linearInfo.bytes = pColInfo->info.bytes;
+    taosArrayPush(pInfo->pLinearInfo, &linearInfo);
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static int32_t initKeeperInfo(STimeSliceOperatorInfo* pInfo, SSDataBlock* pBlock) {
+  int32_t code;
+  code = initPrevRowsKeeper(pInfo, pBlock);
+  if (code != TSDB_CODE_SUCCESS) {
+    return TSDB_CODE_FAILED;
+  }
+
+  code = initNextRowsKeeper(pInfo, pBlock);
+  if (code != TSDB_CODE_SUCCESS) {
+    return TSDB_CODE_FAILED;
+  }
+
+  code = initFillLinearInfo(pInfo, pBlock);
+  if (code != TSDB_CODE_SUCCESS) {
+    return TSDB_CODE_FAILED;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
 static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
  if (pOperator->status == OP_EXEC_DONE) {
    return NULL;
@ -2278,13 +2344,7 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
      break;
    }

-    int32_t code;
-    code = initPrevRowsKeeper(pSliceInfo, pBlock);
-    if (code != TSDB_CODE_SUCCESS) {
-      longjmp(pTaskInfo->env, code);
-    }
-
-    code = initNextRowsKeeper(pSliceInfo, pBlock);
+    int32_t code = initKeeperInfo(pSliceInfo, pBlock);
    if (code != TSDB_CODE_SUCCESS) {
      longjmp(pTaskInfo->env, code);
    }
@ -2312,6 +2372,32 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
        pResBlock->info.rows += 1;
        doKeepPrevRows(pSliceInfo, pBlock, i);

+        // for linear interpolation, always fill value between this and next points;
+        // if its the first point in data block, also fill values between previous(if there's any) and this point;
+        // if its the last point in data block, no need to fill, but reserve this point as the start value for next data block.
+        if (pSliceInfo->fillType == TSDB_FILL_LINEAR) {
+          doKeepLinearInfo(pSliceInfo, pBlock, i);
+          pSliceInfo->current =
+              taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+          if (i < pBlock->info.rows - 1) {
+            int64_t nextTs = *(int64_t*)colDataGetData(pTsCol, i + 1);
+            if (nextTs > pSliceInfo->current) {
+              while (pSliceInfo->current < nextTs && pSliceInfo->current <= pSliceInfo->win.ekey) {
+                genInterpolationResult(pSliceInfo, &pOperator->exprSupp, pBlock, pResBlock);
+                pSliceInfo->current =
+                    taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+                if (pResBlock->info.rows >= pResBlock->info.capacity) {
+                  break;
+                }
+              }
+
+              if (pSliceInfo->current > pSliceInfo->win.ekey) {
+                doSetOperatorCompleted(pOperator);
+                break;
+              }
+            }
+          }
+        } else { // non-linear interpolation
          pSliceInfo->current =
              taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
          if (pSliceInfo->current > pSliceInfo->win.ekey) {
@ -2322,12 +2408,37 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
          if (pResBlock->info.rows >= pResBlock->info.capacity) {
            break;
          }
+        }
      } else if (ts < pSliceInfo->current) {
-        // in case interpolation window starts and ends between two datapoints, fill(prev) need to interpolate
+        // in case of interpolation window starts and ends between two datapoints, fill(prev) need to interpolate
        doKeepPrevRows(pSliceInfo, pBlock, i);

+        if (pSliceInfo->fillType == TSDB_FILL_LINEAR) {
+          doKeepLinearInfo(pSliceInfo, pBlock, i);
+          // no need to increate pSliceInfo->current here
+          //pSliceInfo->current =
+          //    taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
          if (i < pBlock->info.rows - 1) {
-          // in case interpolation window starts and ends between two datapoints, fill(next) need to interpolate
+            int64_t nextTs = *(int64_t*)colDataGetData(pTsCol, i + 1);
+            if (nextTs > pSliceInfo->current) {
+              while (pSliceInfo->current < nextTs && pSliceInfo->current <= pSliceInfo->win.ekey) {
+                genInterpolationResult(pSliceInfo, &pOperator->exprSupp, pBlock, pResBlock);
+                pSliceInfo->current =
+                    taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+                if (pResBlock->info.rows >= pResBlock->info.capacity) {
+                  break;
+                }
+              }
+
+              if (pSliceInfo->current > pSliceInfo->win.ekey) {
+                doSetOperatorCompleted(pOperator);
+                break;
+              }
+            }
+          }
+        } else { // non-linear interpolation
+          if (i < pBlock->info.rows - 1) {
+            // in case of interpolation window starts and ends between two datapoints, fill(next) need to interpolate
            doKeepNextRows(pSliceInfo, pBlock, i + 1);
            int64_t nextTs = *(int64_t*)colDataGetData(pTsCol, i + 1);
            if (nextTs > pSliceInfo->current) {
@ -2350,8 +2461,9 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
          } else {  // it is the last row of current block
            doKeepPrevRows(pSliceInfo, pBlock, i);
          }
+        }
      } else {  // ts > pSliceInfo->current
-        // in case interpolation window starts and ends between two datapoints, fill(next) need to interpolate
+        // in case of interpolation window starts and ends between two datapoints, fill(next) need to interpolate
        doKeepNextRows(pSliceInfo, pBlock, i);

        while (pSliceInfo->current < ts && pSliceInfo->current <= pSliceInfo->win.ekey) {
@ -2380,9 +2492,18 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
          pResBlock->info.rows += 1;
          doKeepPrevRows(pSliceInfo, pBlock, i);

+
+          if (pSliceInfo->fillType == TSDB_FILL_LINEAR) {
+            doKeepLinearInfo(pSliceInfo, pBlock, i);
+            pSliceInfo->current =
+                taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+            if (i < pBlock->info.rows - 1) {
+              int64_t nextTs = *(int64_t*)colDataGetData(pTsCol, i + 1);
+              if (nextTs > pSliceInfo->current) {
+                while (pSliceInfo->current < nextTs && pSliceInfo->current <= pSliceInfo->win.ekey) {
+                  genInterpolationResult(pSliceInfo, &pOperator->exprSupp, pBlock, pResBlock);
                  pSliceInfo->current =
                      taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
-
                  if (pResBlock->info.rows >= pResBlock->info.capacity) {
                    break;
                  }
@ -2394,10 +2515,26 @@ static SSDataBlock* doTimeslice(SOperatorInfo* pOperator) {
                }
              }
            }
+          } else { // non-linear interpolation
+            pSliceInfo->current =
+                taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
+
+            if (pResBlock->info.rows >= pResBlock->info.capacity) {
+              break;
+            }
+          }
+        }
+
+        if (pSliceInfo->current > pSliceInfo->win.ekey) {
+          doSetOperatorCompleted(pOperator);
+          break;
+        }
+      }
+    }

    // check if need to interpolate after ts range
-    // except for fill(next)
-    while (pSliceInfo->current <= pSliceInfo->win.ekey && pSliceInfo->fillType != TSDB_FILL_NEXT) {
+    // except for fill(next), fill(linear)
+    while (pSliceInfo->current <= pSliceInfo->win.ekey && pSliceInfo->fillType != TSDB_FILL_NEXT && pSliceInfo->fillType != TSDB_FILL_LINEAR) {
      genInterpolationResult(pSliceInfo, &pOperator->exprSupp, pBlock, pResBlock);
      pSliceInfo->current =
          taosTimeAdd(pSliceInfo->current, pInterval->interval, pInterval->intervalUnit, pInterval->precision);
@ -2446,6 +2583,9 @@ SOperatorInfo* createTimeSliceOperatorInfo(SOperatorInfo* downstream, SPhysiNode
  pInfo->fillType = convertFillType(pInterpPhyNode->fillMode);
  initResultSizeInfo(&pOperator->resultInfo, 4096);

+  pInfo->pPrevRow = NULL;
+  pInfo->pNextRow = NULL;
+  pInfo->pLinearInfo = NULL;
  pInfo->pFillColInfo = createFillColInfo(pExprInfo, numOfExprs, (SNodeListNode*)pInterpPhyNode->pFillValues);
  pInfo->pRes = createResDataBlock(pPhyNode->pOutputDataBlockDesc);
  pInfo->win = pInterpPhyNode->timeRange;
--- a/source/libs/parser/src/parAstCreater.c
+++ b/source/libs/parser/src/parAstCreater.c
@ -819,6 +819,10 @@ SNode* createDefaultDatabaseOptions(SAstCreateContext* pCxt) {
  pOptions->numOfVgroups = TSDB_DEFAULT_VN_PER_DB;
  pOptions->singleStable = TSDB_DEFAULT_DB_SINGLE_STABLE;
  pOptions->schemaless = TSDB_DEFAULT_DB_SCHEMALESS;
+  pOptions->walRetentionPeriod = TSDB_DEFAULT_DB_WAL_RETENTION_PERIOD;
+  pOptions->walRetentionSize = TSDB_DEFAULT_DB_WAL_RETENTION_SIZE;
+  pOptions->walRollPeriod = TSDB_DEFAULT_DB_WAL_ROLL_PERIOD;
+  pOptions->walSegmentSize = TSDB_DEFAULT_DB_WAL_SEGMENT_SIZE;
  return (SNode*)pOptions;
 }

@ -846,6 +850,10 @@ SNode* createAlterDatabaseOptions(SAstCreateContext* pCxt) {
  pOptions->numOfVgroups = -1;
  pOptions->singleStable = -1;
  pOptions->schemaless = -1;
+  pOptions->walRetentionPeriod = -1;
+  pOptions->walRetentionSize = -1;
+  pOptions->walRollPeriod = -1;
+  pOptions->walSegmentSize = -1;
  return (SNode*)pOptions;
 }

--- a/source/libs/parser/src/parInsert.c
+++ b/source/libs/parser/src/parInsert.c
@ -1537,6 +1537,13 @@ int32_t parseInsertSql(SParseContext* pContext, SQuery** pQuery, SParseMetaCache
  if (pContext->pStmtCb && *pQuery) {
    (*pContext->pStmtCb->getExecInfoFn)(pContext->pStmtCb->pStmt, &context.pVgroupsHashObj,
                                        &context.pTableBlockHashObj);
+    if (NULL == context.pVgroupsHashObj) {
+      context.pVgroupsHashObj = taosHashInit(128, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
+    }
+    if (NULL == context.pTableBlockHashObj) {
+      context.pTableBlockHashObj =
+          taosHashInit(128, taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY), true, HASH_NO_LOCK);
+    }
  } else {
    context.pVgroupsHashObj = taosHashInit(128, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), true, HASH_NO_LOCK);
    context.pTableBlockHashObj =
--- a/source/libs/parser/src/parInsertData.c
+++ b/source/libs/parser/src/parInsertData.c
@ -525,7 +525,7 @@ static int sortMergeDataBlockDupRows(STableDataBlocks* dataBuf, SBlockKeyInfo* p
                                     SBlockRowMerger** ppBlkRowMerger) {
  SSubmitBlk* pBlocks = (SSubmitBlk*)dataBuf->pData;
  STableMeta* pTableMeta = dataBuf->pTableMeta;
-  int16_t     nRows = pBlocks->numOfRows;
+  int32_t     nRows = pBlocks->numOfRows;

  // size is less than the total size, since duplicated rows may be removed.

@ -546,7 +546,7 @@ static int sortMergeDataBlockDupRows(STableDataBlocks* dataBuf, SBlockKeyInfo* p
  int32_t         extendedRowSize = getExtendedRowSize(dataBuf);
  SBlockKeyTuple* pBlkKeyTuple = pBlkKeyInfo->pKeyTuple;
  char*           pBlockData = pBlocks->data + pBlocks->schemaLen;
-  int             n = 0;
+  int32_t         n = 0;
  while (n < nRows) {
    pBlkKeyTuple->skey = TD_ROW_KEY((STSRow*)pBlockData);
    pBlkKeyTuple->payloadAddr = pBlockData;
--- a/source/libs/parser/test/parInitialCTest.cpp
+++ b/source/libs/parser/test/parInitialCTest.cpp
@ -111,6 +111,10 @@ TEST_F(ParserInitialCTest, createDatabase) {
    expect.numOfVgroups = TSDB_DEFAULT_VN_PER_DB;
    expect.numOfStables = TSDB_DEFAULT_DB_SINGLE_STABLE;
    expect.schemaless = TSDB_DEFAULT_DB_SCHEMALESS;
+    expect.walRetentionPeriod = TSDB_DEFAULT_DB_WAL_RETENTION_PERIOD;
+    expect.walRetentionSize = TSDB_DEFAULT_DB_WAL_RETENTION_SIZE;
+    expect.walRollPeriod = TSDB_DEFAULT_DB_WAL_ROLL_PERIOD;
+    expect.walSegmentSize = TSDB_DEFAULT_DB_WAL_SEGMENT_SIZE;
  };

  auto setDbBufferFunc = [&](int32_t buffer) { expect.buffer = buffer; };
--- a/source/libs/scheduler/inc/schInt.h
+++ b/source/libs/scheduler/inc/schInt.h
@ -61,6 +61,8 @@ typedef enum {
 #define SCH_MAX_TASK_TIMEOUT_USEC 60000000
 #define SCH_DEFAULT_MAX_RETRY_NUM 6

+#define SCH_ASYNC_LAUNCH_TASK 0
+
 typedef struct SSchDebug {
  bool     lockEnable;
  bool     apiEnable;
@ -281,6 +283,11 @@ typedef struct SSchJob {
  SQueryProfileSummary summary;
 } SSchJob;

+typedef struct SSchTaskCtx {
+  SSchJob  *pJob;
+  SSchTask *pTask;
+} SSchTaskCtx;
+
 extern SSchedulerMgmt schMgmt;

 #define SCH_TASK_TIMEOUT(_task) ((taosGetTimestampUs() - *(int64_t*)taosArrayGet((_task)->profile.execTime, (_task)->execId)) > (_task)->timeoutUsec)
@ -428,7 +435,7 @@ int32_t schChkJobNeedFlowCtrl(SSchJob *pJob, SSchLevel *pLevel);
 int32_t schDecTaskFlowQuota(SSchJob *pJob, SSchTask *pTask);
 int32_t schCheckIncTaskFlowQuota(SSchJob *pJob, SSchTask *pTask, bool *enough);
 int32_t schLaunchTasksInFlowCtrlList(SSchJob *pJob, SSchTask *pTask);
-int32_t schLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask);
+int32_t schAsyncLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask);
 int32_t schLaunchFetchTask(SSchJob *pJob);
 int32_t schProcessOnTaskFailure(SSchJob *pJob, SSchTask *pTask, int32_t errCode);
 int32_t schBuildAndSendHbMsg(SQueryNodeEpId *nodeEpId, SArray* taskAction);
--- a/source/libs/scheduler/src/schFlowCtrl.c
+++ b/source/libs/scheduler/src/schFlowCtrl.c
@ -54,7 +54,7 @@ int32_t schChkJobNeedFlowCtrl(SSchJob *pJob, SSchLevel *pLevel) {
    sum += pTask->plan->execNodeStat.tableNum;
  }

-  if (sum < schMgmt.cfg.maxNodeTableNum) {
+  if (schMgmt.cfg.maxNodeTableNum <= 0 || sum < schMgmt.cfg.maxNodeTableNum) {
    SCH_JOB_DLOG("job no need flow ctrl, totalTableNum:%d", sum);
    return TSDB_CODE_SUCCESS;
  }
@ -230,7 +230,7 @@ int32_t schLaunchTasksInFlowCtrlListImpl(SSchJob *pJob, SSchFlowControl *ctrl) {
    SCH_TASK_DLOG("task to launch, fqdn:%s, port:%d, tableNum:%d, remainNum:%d, remainExecTaskNum:%d", 
       ep->fqdn, ep->port, pTask->plan->execNodeStat.tableNum, ctrl->tableNumSum, ctrl->execTaskNum);
    
-    SCH_ERR_JRET(schLaunchTaskImpl(pJob, pTask));
+    SCH_ERR_JRET(schAsyncLaunchTaskImpl(pJob, pTask));
    
    remainNum -= pTask->plan->execNodeStat.tableNum;
    if (remainNum <= 0) {
--- a/source/libs/scheduler/src/schTask.c
+++ b/source/libs/scheduler/src/schTask.c
@ -411,7 +411,7 @@ int32_t schHandleRedirect(SSchJob *pJob, SSchTask *pTask, SDataBuf *pData, int32
    if (pJob->fetched) {
      SCH_UNLOCK(SCH_WRITE, &pJob->resLock);
      SCH_TASK_ELOG("already fetched while got error %s", tstrerror(rspCode));
-      SCH_ERR_RET(rspCode);
+      SCH_ERR_JRET(rspCode);
    }
    SCH_UNLOCK(SCH_WRITE, &pJob->resLock);

@ -819,7 +819,10 @@ int32_t schProcessOnTaskStatusRsp(SQueryNodeEpId *pEpId, SArray *pStatusList) {
  return TSDB_CODE_SUCCESS;
 }

-int32_t schLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask) {
+int32_t schLaunchTaskImpl(void *param) {
+  SSchTaskCtx *pCtx = (SSchTaskCtx *)param;
+  SSchJob *pJob = pCtx->pJob;
+  SSchTask *pTask = pCtx->pTask;
  int8_t  status = 0;
  int32_t code = 0;

@ -834,12 +837,12 @@ int32_t schLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask) {

  if (schJobNeedToStop(pJob, &status)) {
    SCH_TASK_DLOG("no need to launch task cause of job status %s", jobTaskStatusStr(status));
-    SCH_ERR_RET(TSDB_CODE_SCH_IGNORE_ERROR);
+    SCH_ERR_JRET(TSDB_CODE_SCH_IGNORE_ERROR);
  }

  // NOTE: race condition: the task should be put into the hash table before send msg to server
  if (SCH_GET_TASK_STATUS(pTask) != JOB_TASK_STATUS_EXEC) {
-    SCH_ERR_RET(schPushTaskToExecList(pJob, pTask));
+    SCH_ERR_JRET(schPushTaskToExecList(pJob, pTask));
    SCH_SET_TASK_STATUS(pTask, JOB_TASK_STATUS_EXEC);
  }

@ -850,19 +853,51 @@ int32_t schLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask) {
    if (TSDB_CODE_SUCCESS != code) {
      SCH_TASK_ELOG("failed to create physical plan, code:%s, msg:%p, len:%d", tstrerror(code), pTask->msg,
                    pTask->msgLen);
-      SCH_ERR_RET(code);
+      SCH_ERR_JRET(code);
    } else {
      SCH_TASK_DLOGL("physical plan len:%d, %s", pTask->msgLen, pTask->msg);
    }
  }

-  SCH_ERR_RET(schSetTaskCandidateAddrs(pJob, pTask));
+  SCH_ERR_JRET(schSetTaskCandidateAddrs(pJob, pTask));

  if (SCH_IS_QUERY_JOB(pJob)) {
-    SCH_ERR_RET(schEnsureHbConnection(pJob, pTask));
+    SCH_ERR_JRET(schEnsureHbConnection(pJob, pTask));
  }

-  SCH_ERR_RET(schBuildAndSendMsg(pJob, pTask, NULL, plan->msgType));
+  SCH_ERR_JRET(schBuildAndSendMsg(pJob, pTask, NULL, plan->msgType));
+
+_return:
+
+  taosMemoryFree(param);
+
+#if SCH_ASYNC_LAUNCH_TASK
+  if (code) {
+    code = schProcessOnTaskFailure(pJob, pTask, code);
+  }
+  if (code) {
+    code = schHandleJobFailure(pJob, code);
+  }
+#endif
+
+  SCH_RET(code);
+}
+
+int32_t schAsyncLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask) {  
+
+  SSchTaskCtx *param = taosMemoryCalloc(1, sizeof(SSchTaskCtx));
+  if (NULL == param) {
+    SCH_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
+  }
+  
+  param->pJob = pJob;
+  param->pTask = pTask;
+
+#if SCH_ASYNC_LAUNCH_TASK
+  taosAsyncExec(schLaunchTaskImpl, param, NULL);
+#else
+  SCH_ERR_RET(schLaunchTaskImpl(param));
+#endif

  return TSDB_CODE_SUCCESS;
 }
@ -878,10 +913,10 @@ int32_t schLaunchTask(SSchJob *pJob, SSchTask *pTask) {
    SCH_ERR_JRET(schCheckIncTaskFlowQuota(pJob, pTask, &enough));

    if (enough) {
-      SCH_ERR_JRET(schLaunchTaskImpl(pJob, pTask));
+      SCH_ERR_JRET(schAsyncLaunchTaskImpl(pJob, pTask));
    }
  } else {
-    SCH_ERR_JRET(schLaunchTaskImpl(pJob, pTask));
+    SCH_ERR_JRET(schAsyncLaunchTaskImpl(pJob, pTask));
  }

  return TSDB_CODE_SUCCESS;
--- a/source/libs/scheduler/src/scheduler.c
+++ b/source/libs/scheduler/src/scheduler.c
@ -154,6 +154,8 @@ void schedulerFreeJob(int64_t* jobId, int32_t errCode) {
    return;
  }

+  SCH_JOB_DLOG("start to free job 0x%" PRIx64 ", errCode:0x%x", *jobId, errCode);
+
  schHandleJobDrop(pJob, errCode);
  
  schReleaseJob(*jobId);
--- a/source/libs/stream/src/stream.c
+++ b/source/libs/stream/src/stream.c
@ -136,6 +136,7 @@ int32_t streamTaskEnqueue(SStreamTask* pTask, SStreamDispatchReq* pReq, SRpcMsg*
  pRsp->pCont = buf;
  pRsp->contLen = sizeof(SMsgHead) + sizeof(SStreamDispatchRsp);
  tmsgSendRsp(pRsp);
+  tFreeStreamDispatchReq(pReq);
  return status == TASK_INPUT_STATUS__NORMAL ? 0 : -1;
 }

@ -228,6 +229,7 @@ int32_t streamProcessRunReq(SStreamTask* pTask) {
  return 0;
 }

+#if 0
 int32_t streamProcessRecoverReq(SStreamTask* pTask, SStreamTaskRecoverReq* pReq, SRpcMsg* pRsp) {
  void* buf = rpcMallocCont(sizeof(SMsgHead) + sizeof(SStreamTaskRecoverRsp));
  ((SMsgHead*)buf)->vgId = htonl(pReq->upstreamNodeId);
@ -266,6 +268,7 @@ int32_t streamProcessRecoverRsp(SStreamTask* pTask, SStreamTaskRecoverRsp* pRsp)

  return 0;
 }
+#endif

 int32_t streamProcessRetrieveReq(SStreamTask* pTask, SStreamRetrieveReq* pReq, SRpcMsg* pRsp) {
  qDebug("task %d receive retrieve req from node %d task %d", pTask->taskId, pReq->srcNodeId, pReq->srcTaskId);
--- a/source/libs/stream/src/streamDispatch.c
+++ b/source/libs/stream/src/streamDispatch.c
@ -62,6 +62,11 @@ int32_t tDecodeStreamDispatchReq(SDecoder* pDecoder, SStreamDispatchReq* pReq) {
  return 0;
 }

+void tFreeStreamDispatchReq(SStreamDispatchReq* pReq) {
+  taosArrayDestroyP(pReq->data, taosMemoryFree);
+  taosArrayDestroy(pReq->dataLen);
+}
+
 int32_t tEncodeStreamRetrieveReq(SEncoder* pEncoder, const SStreamRetrieveReq* pReq) {
  if (tStartEncode(pEncoder) < 0) return -1;
  if (tEncodeI64(pEncoder, pReq->streamId) < 0) return -1;
@ -279,7 +284,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
    }
    code = 0;
  FAIL_FIXED_DISPATCH:
-    taosArrayDestroy(req.data);
+    taosArrayDestroyP(req.data, taosMemoryFree);
    taosArrayDestroy(req.dataLen);
    return code;

--- a/source/libs/stream/src/streamExec.c
+++ b/source/libs/stream/src/streamExec.c
@ -15,26 +15,26 @@

 #include "streamInc.h"

-static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes) {
+static int32_t streamTaskExecImpl(SStreamTask* pTask, const void* data, SArray* pRes) {
  void* exec = pTask->exec.executor;

  // set input
-  SStreamQueueItem* pItem = (SStreamQueueItem*)data;
+  const SStreamQueueItem* pItem = (const SStreamQueueItem*)data;
  if (pItem->type == STREAM_INPUT__GET_RES) {
-    SStreamTrigger* pTrigger = (SStreamTrigger*)data;
+    const SStreamTrigger* pTrigger = (const SStreamTrigger*)data;
    qSetMultiStreamInput(exec, pTrigger->pBlock, 1, STREAM_INPUT__DATA_BLOCK);
  } else if (pItem->type == STREAM_INPUT__DATA_SUBMIT) {
    ASSERT(pTask->taskLevel == TASK_LEVEL__SOURCE);
-    SStreamDataSubmit* pSubmit = (SStreamDataSubmit*)data;
+    const SStreamDataSubmit* pSubmit = (const SStreamDataSubmit*)data;
    qDebug("task %d %p set submit input %p %p %d 1", pTask->taskId, pTask, pSubmit, pSubmit->data, *pSubmit->dataRef);
    qSetMultiStreamInput(exec, pSubmit->data, 1, STREAM_INPUT__DATA_SUBMIT);
  } else if (pItem->type == STREAM_INPUT__DATA_BLOCK || pItem->type == STREAM_INPUT__DATA_RETRIEVE) {
-    SStreamDataBlock* pBlock = (SStreamDataBlock*)data;
+    const SStreamDataBlock* pBlock = (const SStreamDataBlock*)data;
    SArray*                 blocks = pBlock->blocks;
    qDebug("task %d %p set ssdata input", pTask->taskId, pTask);
    qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_INPUT__DATA_BLOCK);
  } else if (pItem->type == STREAM_INPUT__MERGED_SUBMIT) {
-    SStreamMergedSubmit* pMerged = (SStreamMergedSubmit*)data;
+    const SStreamMergedSubmit* pMerged = (const SStreamMergedSubmit*)data;
    SArray*                    blocks = pMerged->reqs;
    qDebug("task %d %p set submit input (merged), batch num: %d", pTask->taskId, pTask, (int32_t)blocks->size);
    qSetMultiStreamInput(exec, blocks->pData, blocks->size, STREAM_INPUT__MERGED_SUBMIT);
@ -52,7 +52,7 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
    if (output == NULL) {
      if (pItem->type == STREAM_INPUT__DATA_RETRIEVE) {
        SSDataBlock             block = {0};
-        SStreamDataBlock* pRetrieveBlock = (SStreamDataBlock*)data;
+        const SStreamDataBlock* pRetrieveBlock = (const SStreamDataBlock*)data;
        ASSERT(taosArrayGetSize(pRetrieveBlock->blocks) == 1);
        assignOneDataBlock(&block, taosArrayGet(pRetrieveBlock->blocks, 0));
        block.info.type = STREAM_PULL_OVER;
@ -82,14 +82,16 @@ static int32_t streamTaskExecImpl(SStreamTask* pTask, void* data, SArray* pRes)
  return 0;
 }

+#if 0
 static FORCE_INLINE int32_t streamUpdateVer(SStreamTask* pTask, SStreamDataBlock* pBlock) {
  ASSERT(pBlock->type == STREAM_INPUT__DATA_BLOCK);
  int32_t             childId = pBlock->childId;
  int64_t             ver = pBlock->sourceVer;
  SStreamChildEpInfo* pChildInfo = taosArrayGetP(pTask->childEpInfo, childId);
-  pChildInfo->processedVer = ver;
+  /*pChildInfo-> = ver;*/
  return 0;
 }
+#endif

 int32_t streamPipelineExec(SStreamTask* pTask, int32_t batchNum) {
  ASSERT(pTask->taskLevel != TASK_LEVEL__SINK);
@ -204,6 +206,7 @@ int32_t streamExecForAll(SStreamTask* pTask) {
        // TODO log failed ver
        streamQueueProcessFail(pTask->inputQueue);
        taosArrayDestroy(pRes);
+        streamFreeQitem(data);
        return -1;
      }
      qRes->type = STREAM_INPUT__DATA_BLOCK;
@ -220,10 +223,14 @@ int32_t streamExecForAll(SStreamTask* pTask) {
        /*streamQueueProcessFail(pTask->inputQueue);*/
        taosArrayDestroyEx(pRes, (FDelete)blockDataFreeRes);
        taosFreeQitem(qRes);
+        streamFreeQitem(data);
        return -1;
      }
      /*streamQueueProcessSuccess(pTask->inputQueue);*/
+    } else {
+      taosArrayDestroy(pRes);
    }
+    streamFreeQitem(data);
  }
  return 0;
 }
--- a/source/libs/stream/src/streamMeta.c
+++ b/source/libs/stream/src/streamMeta.c
@ -48,8 +48,18 @@ SStreamMeta* streamMetaOpen(const char* path, void* ahandle, FTaskExpand expandF
  pMeta->ahandle = ahandle;
  pMeta->expandFunc = expandFunc;

+  if (streamLoadTasks(pMeta) < 0) {
+    goto _err;
+  }
  return pMeta;
+
 _err:
+  if (pMeta->path) taosMemoryFree(pMeta->path);
+  if (pMeta->pTasks) taosHashCleanup(pMeta->pTasks);
+  if (pMeta->pStateDb) tdbTbClose(pMeta->pStateDb);
+  if (pMeta->pTaskDb) tdbTbClose(pMeta->pTaskDb);
+  if (pMeta->db) tdbClose(pMeta->db);
+  taosMemoryFree(pMeta);
  return NULL;
 }

--- a/source/libs/stream/src/streamRecover.c
+++ b/source/libs/stream/src/streamRecover.c
@ -15,6 +15,7 @@

 #include "streamInc.h"

+#if 0
 int32_t tEncodeStreamTaskRecoverReq(SEncoder* pEncoder, const SStreamTaskRecoverReq* pReq) {
  if (tStartEncode(pEncoder) < 0) return -1;
  if (tEncodeI64(pEncoder, pReq->streamId) < 0) return -1;
@ -86,67 +87,104 @@ int32_t tDecodeSMStreamTaskRecoverRsp(SDecoder* pDecoder, SMStreamTaskRecoverRsp
  tEndDecode(pDecoder);
  return 0;
 }
+#endif

-typedef struct {
-  int32_t vgId;
-  int32_t childId;
-  int64_t ver;
-} SStreamVgVerCheckpoint;
-
-int32_t tEncodeSStreamVgVerCheckpoint(SEncoder* pEncoder, const SStreamVgVerCheckpoint* pCheckpoint) {
-  if (tEncodeI32(pEncoder, pCheckpoint->vgId) < 0) return -1;
-  if (tEncodeI32(pEncoder, pCheckpoint->childId) < 0) return -1;
-  if (tEncodeI64(pEncoder, pCheckpoint->ver) < 0) return -1;
+int32_t tEncodeSStreamCheckpointInfo(SEncoder* pEncoder, const SStreamCheckpointInfo* pCheckpoint) {
+  if (tEncodeI32(pEncoder, pCheckpoint->srcNodeId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pCheckpoint->srcChildId) < 0) return -1;
+  if (tEncodeI64(pEncoder, pCheckpoint->stateProcessedVer) < 0) return -1;
  return 0;
 }

-int32_t tDecodeSStreamVgVerCheckpoint(SDecoder* pDecoder, SStreamVgVerCheckpoint* pCheckpoint) {
-  if (tDecodeI32(pDecoder, &pCheckpoint->vgId) < 0) return -1;
-  if (tDecodeI32(pDecoder, &pCheckpoint->childId) < 0) return -1;
-  if (tDecodeI64(pDecoder, &pCheckpoint->ver) < 0) return -1;
+int32_t tDecodeSStreamCheckpointInfo(SDecoder* pDecoder, SStreamCheckpointInfo* pCheckpoint) {
+  if (tDecodeI32(pDecoder, &pCheckpoint->srcNodeId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pCheckpoint->srcChildId) < 0) return -1;
+  if (tDecodeI64(pDecoder, &pCheckpoint->stateProcessedVer) < 0) return -1;
  return 0;
 }

-typedef struct {
-  int64_t streamId;
-  int64_t checkTs;
-  int64_t checkpointId;
-  int32_t taskId;
-  SArray* checkpointVer;  // SArray<SStreamVgCheckpointVer>
-} SStreamAggVerCheckpoint;
-
-int32_t tEncodeSStreamAggVerCheckpoint(SEncoder* pEncoder, const SStreamAggVerCheckpoint* pCheckpoint) {
+int32_t tEncodeSStreamMultiVgCheckpointInfo(SEncoder* pEncoder, const SStreamMultiVgCheckpointInfo* pCheckpoint) {
  if (tEncodeI64(pEncoder, pCheckpoint->streamId) < 0) return -1;
  if (tEncodeI64(pEncoder, pCheckpoint->checkTs) < 0) return -1;
-  if (tEncodeI64(pEncoder, pCheckpoint->checkpointId) < 0) return -1;
+  if (tEncodeI32(pEncoder, pCheckpoint->checkpointId) < 0) return -1;
  if (tEncodeI32(pEncoder, pCheckpoint->taskId) < 0) return -1;
  int32_t sz = taosArrayGetSize(pCheckpoint->checkpointVer);
  if (tEncodeI32(pEncoder, sz) < 0) return -1;
  for (int32_t i = 0; i < sz; i++) {
-    SStreamVgVerCheckpoint* pOneVgCkpoint = taosArrayGet(pCheckpoint->checkpointVer, i);
-    if (tEncodeSStreamVgVerCheckpoint(pEncoder, pOneVgCkpoint) < 0) return -1;
+    SStreamCheckpointInfo* pOneVgCkpoint = taosArrayGet(pCheckpoint->checkpointVer, i);
+    if (tEncodeSStreamCheckpointInfo(pEncoder, pOneVgCkpoint) < 0) return -1;
  }
  return 0;
 }

-int32_t tDecodeSStreamAggVerCheckpoint(SDecoder* pDecoder, SStreamAggVerCheckpoint* pCheckpoint) {
+int32_t tDecodeSStreamMultiVgCheckpointInfo(SDecoder* pDecoder, SStreamMultiVgCheckpointInfo* pCheckpoint) {
  if (tDecodeI64(pDecoder, &pCheckpoint->streamId) < 0) return -1;
  if (tDecodeI64(pDecoder, &pCheckpoint->checkTs) < 0) return -1;
-  if (tDecodeI64(pDecoder, &pCheckpoint->checkpointId) < 0) return -1;
+  if (tDecodeI32(pDecoder, &pCheckpoint->checkpointId) < 0) return -1;
  if (tDecodeI32(pDecoder, &pCheckpoint->taskId) < 0) return -1;
  int32_t sz;
  if (tDecodeI32(pDecoder, &sz) < 0) return -1;
  for (int32_t i = 0; i < sz; i++) {
-    SStreamVgVerCheckpoint oneVgCheckpoint;
-    if (tDecodeSStreamVgVerCheckpoint(pDecoder, &oneVgCheckpoint) < 0) return -1;
+    SStreamCheckpointInfo oneVgCheckpoint;
+    if (tDecodeSStreamCheckpointInfo(pDecoder, &oneVgCheckpoint) < 0) return -1;
    taosArrayPush(pCheckpoint->checkpointVer, &oneVgCheckpoint);
  }
  return 0;
 }

-int32_t streamRecoverSinkLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+int32_t streamSaveStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
+  void* buf = NULL;
+
  ASSERT(pTask->taskLevel == TASK_LEVEL__SINK);
-  // load status
+
+  SStreamMultiVgCheckpointInfo checkpoint;
+  checkpoint.checkpointId = atomic_fetch_add_32(&pTask->nextCheckId, 1);
+  checkpoint.checkTs = taosGetTimestampMs();
+  checkpoint.streamId = pTask->streamId;
+  checkpoint.taskId = pTask->taskId;
+  checkpoint.checkpointVer = pTask->checkpointInfo;
+
+  int32_t len;
+  int32_t code;
+  tEncodeSize(tEncodeSStreamMultiVgCheckpointInfo, &checkpoint, len, code);
+  if (code < 0) {
+    return -1;
+  }
+
+  buf = taosMemoryCalloc(1, len);
+  if (buf == NULL) {
+    return -1;
+  }
+  SEncoder encoder;
+  tEncoderInit(&encoder, buf, len);
+  tEncodeSStreamMultiVgCheckpointInfo(&encoder, &checkpoint);
+  tEncoderClear(&encoder);
+
+  SStreamCheckpointKey key = {
+      .taskId = pTask->taskId,
+      .checkpointId = checkpoint.checkpointId,
+  };
+
+  if (tdbTbUpsert(pMeta->pStateDb, &key, sizeof(SStreamCheckpointKey), buf, len, &pMeta->txn) < 0) {
+    ASSERT(0);
+    goto FAIL;
+  }
+
+  int32_t sz = taosArrayGetSize(pTask->checkpointInfo);
+  for (int32_t i = 0; i < sz; i++) {
+    SStreamCheckpointInfo* pCheck = taosArrayGet(pTask->checkpointInfo, i);
+    pCheck->stateSaveVer = pCheck->stateProcessedVer;
+  }
+
+  taosMemoryFree(buf);
+  return 0;
+FAIL:
+  if (buf) taosMemoryFree(buf);
+  return -1;
+  return 0;
+}
+
+int32_t streamLoadStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
  void*   pVal = NULL;
  int32_t vLen = 0;
  if (tdbTbGet(pMeta->pStateDb, &pTask->taskId, sizeof(void*), &pVal, &vLen) < 0) {
@ -154,9 +192,87 @@ int32_t streamRecoverSinkLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
  }
  SDecoder decoder;
  tDecoderInit(&decoder, pVal, vLen);
-  SStreamAggVerCheckpoint aggCheckpoint;
-  tDecodeSStreamAggVerCheckpoint(&decoder, &aggCheckpoint);
-  /*pTask->*/
+  SStreamMultiVgCheckpointInfo aggCheckpoint;
+  tDecodeSStreamMultiVgCheckpointInfo(&decoder, &aggCheckpoint);
+  tDecoderClear(&decoder);
+
+  pTask->nextCheckId = aggCheckpoint.checkpointId + 1;
+  pTask->checkpointInfo = aggCheckpoint.checkpointVer;
+
+  return 0;
+}
+
+int32_t streamSaveSinkLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__SINK);
+  return streamSaveStateInfo(pMeta, pTask);
+}
+
+int32_t streamRecoverSinkLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__SINK);
+  return streamLoadStateInfo(pMeta, pTask);
+}
+
+int32_t streamSaveAggLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__AGG);
+  // TODO save and copy state
+
+  // save state info
+  if (streamSaveStateInfo(pMeta, pTask) < 0) {
+    return -1;
+  }
+  return 0;
+}
+
+int32_t streamFetchDownstreamStatus(SStreamTask* pTask) {
+  // set self status to recover_phase1
+  // build fetch status msg
+  // send fetch msg
+  atomic_store_8(&pTask->taskStatus, TASK_STATUS__RECOVER_DOWNSTREAM);
+
+  if (pTask->outputType == TASK_OUTPUT__FIXED_DISPATCH) {
+  } else if (pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
+  } else {
+    ASSERT(0);
+  }
+  return 0;
+}
+
+int32_t streamProcessFetchStatusRsp(SStreamMeta* pMeta, SStreamTask* pTask, void* msg) {
+  // if failed, set timer and retry
+  // if successful
+  // add rsp state to partial recover hash
+  // if complete, begin actual recover
+  return 0;
+}
+
+int32_t streamRecoverAggLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__AGG);
+  // recover sink level
+  // after all sink level recovered
+  // choose suitable state to recover
+  return 0;
+}
+
+int32_t streamSaveSourceLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__SOURCE);
+  // TODO: save and copy state
+  return 0;
+}
+
+int32_t streamRecoverSourceLevel(SStreamMeta* pMeta, SStreamTask* pTask) {
+  ASSERT(pTask->taskLevel == TASK_LEVEL__SOURCE);
+  // if totLevel == 3
+  // fetch agg state
+  // recover from local state to agg state, not send msg
+  // recover from agg state to most recent log v1
+  // enable input queue, set status recover_phase2
+  // recover from v1 to queue msg v2, set status normal
+
+  // if totLevel == 2
+  // fetch sink state
+  // recover from local state to sink state v1, send msg
+  // enable input queue, set status recover_phase2
+  // recover from v1 to queue msg v2, set status normal
  return 0;
 }

--- a/source/libs/stream/src/streamTask.c
+++ b/source/libs/stream/src/streamTask.c
@ -34,7 +34,7 @@ int32_t tEncodeStreamEpInfo(SEncoder* pEncoder, const SStreamChildEpInfo* pInfo)
  if (tEncodeI32(pEncoder, pInfo->taskId) < 0) return -1;
  if (tEncodeI32(pEncoder, pInfo->nodeId) < 0) return -1;
  if (tEncodeI32(pEncoder, pInfo->childId) < 0) return -1;
-  if (tEncodeI64(pEncoder, pInfo->processedVer) < 0) return -1;
+  /*if (tEncodeI64(pEncoder, pInfo->processedVer) < 0) return -1;*/
  if (tEncodeSEpSet(pEncoder, &pInfo->epSet) < 0) return -1;
  return 0;
 }
@ -43,7 +43,7 @@ int32_t tDecodeStreamEpInfo(SDecoder* pDecoder, SStreamChildEpInfo* pInfo) {
  if (tDecodeI32(pDecoder, &pInfo->taskId) < 0) return -1;
  if (tDecodeI32(pDecoder, &pInfo->nodeId) < 0) return -1;
  if (tDecodeI32(pDecoder, &pInfo->childId) < 0) return -1;
-  if (tDecodeI64(pDecoder, &pInfo->processedVer) < 0) return -1;
+  /*if (tDecodeI64(pDecoder, &pInfo->processedVer) < 0) return -1;*/
  if (tDecodeSEpSet(pDecoder, &pInfo->epSet) < 0) return -1;
  return 0;
 }
--- a/source/libs/stream/src/streamUpdate.c
+++ b/source/libs/stream/src/streamUpdate.c
@ -14,6 +14,7 @@
 */

 #include "tstreamUpdate.h"
+#include "tencode.h"
 #include "ttime.h"
 #include "query.h"

@ -250,3 +251,110 @@ void updateInfoDestoryColseWinSBF(SUpdateInfo *pInfo) {
  tScalableBfDestroy(pInfo->pCloseWinSBF);
  pInfo->pCloseWinSBF = NULL;
 }
+
+int32_t updateInfoSerialize(void *buf, int32_t bufLen, const SUpdateInfo *pInfo) {
+  ASSERT(pInfo);
+  SEncoder encoder = {0};
+  tEncoderInit(&encoder, buf, bufLen);
+  if (tStartEncode(&encoder) < 0) return -1;
+
+  int32_t size = taosArrayGetSize(pInfo->pTsBuckets);
+  if (tEncodeI32(&encoder, size) < 0) return -1;
+  for (int32_t i = 0; i < size; i++) {
+    TSKEY* pTs = (TSKEY*)taosArrayGet(pInfo->pTsBuckets, i);
+    if (tEncodeI64(&encoder, *pTs) < 0) return -1;
+  }
+
+  if (tEncodeU64(&encoder, pInfo->numBuckets) < 0) return -1;
+
+  int32_t sBfSize = taosArrayGetSize(pInfo->pTsSBFs);
+  if (tEncodeI32(&encoder, sBfSize) < 0) return -1;
+  for (int32_t i = 0; i < sBfSize; i++) {
+    SScalableBf* pSBf = taosArrayGetP(pInfo->pTsSBFs, i);
+    if (tScalableBfEncode(pSBf, &encoder) < 0) return -1;
+  }
+
+  if (tEncodeU64(&encoder, pInfo->numSBFs) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->interval) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->watermark) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->minTS) < 0) return -1;
+  
+  if (tScalableBfEncode(pInfo->pCloseWinSBF, &encoder) < 0) return -1;
+
+  int32_t mapSize = taosHashGetSize(pInfo->pMap);
+  if (tEncodeI32(&encoder, mapSize) < 0) return -1;
+  void*  pIte = NULL;
+  size_t keyLen = 0;
+  while ((pIte = taosHashIterate(pInfo->pMap, pIte)) != NULL) {
+    void* key = taosHashGetKey(pIte, &keyLen);
+    if (tEncodeU64(&encoder, *(uint64_t*)key) < 0) return -1;
+    if (tEncodeI64(&encoder, *(TSKEY*)pIte) < 0) return -1;
+  }
+
+  if (tEncodeI64(&encoder, pInfo->scanWindow.skey) < 0) return -1;
+  if (tEncodeI64(&encoder, pInfo->scanWindow.ekey) < 0) return -1;
+  if (tEncodeU64(&encoder, pInfo->scanGroupId) < 0) return -1;
+  if (tEncodeU64(&encoder, pInfo->maxVersion) < 0) return -1;
+
+  tEndEncode(&encoder);
+
+  int32_t tlen = encoder.pos;
+  tEncoderClear(&encoder);
+  return tlen;
+}
+
+int32_t updateInfoDeserialize(void *buf, int32_t bufLen, SUpdateInfo *pInfo) {
+  ASSERT(pInfo);
+  SDecoder decoder = {0};
+  tDecoderInit(&decoder, buf, bufLen);
+  if (tStartDecode(&decoder) < 0) return -1;
+
+  int32_t size = 0;
+  if (tDecodeI32(&decoder, &size) < 0) return -1;
+  pInfo->pTsBuckets =  taosArrayInit(size, sizeof(TSKEY));
+  TSKEY ts = INT64_MIN;
+  for (int32_t i = 0; i < size; i++) {
+    if (tDecodeI64(&decoder, &ts) < 0) return -1;
+    taosArrayPush(pInfo->pTsBuckets, &ts);
+  }
+
+  if (tDecodeU64(&decoder, &pInfo->numBuckets) < 0) return -1;
+
+  int32_t sBfSize = 0;
+  if (tDecodeI32(&decoder, &sBfSize) < 0) return -1;
+  pInfo->pTsSBFs = taosArrayInit(sBfSize, sizeof(void *));
+  for (int32_t i = 0; i < sBfSize; i++) {
+    SScalableBf* pSBf = tScalableBfDecode(&decoder);
+    if (!pSBf) return -1;
+    taosArrayPush(pInfo->pTsSBFs, &pSBf);
+  }
+
+  if (tDecodeU64(&decoder, &pInfo->numSBFs) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->interval) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->watermark) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->minTS) < 0) return -1;
+  pInfo->pCloseWinSBF = tScalableBfDecode(&decoder);
+
+  int32_t mapSize = 0;
+  if (tDecodeI32(&decoder, &mapSize) < 0) return -1;
+  _hash_fn_t hashFn = taosGetDefaultHashFunction(TSDB_DATA_TYPE_BINARY);
+  pInfo->pMap = taosHashInit(mapSize, hashFn, true, HASH_NO_LOCK);
+  uint64_t uid = 0;
+  ts = INT64_MIN;
+  for(int32_t i = 0; i < mapSize; i++) {
+    if (tDecodeU64(&decoder, &uid) < 0) return -1;
+    if (tDecodeI64(&decoder, &ts) < 0) return -1;
+    taosHashPut(pInfo->pMap, &uid, sizeof(uint64_t), &ts, sizeof(TSKEY));
+  }
+  ASSERT(mapSize == taosHashGetSize(pInfo->pMap));
+
+  if (tDecodeI64(&decoder, &pInfo->scanWindow.skey) < 0) return -1;
+  if (tDecodeI64(&decoder, &pInfo->scanWindow.ekey) < 0) return -1;
+  if (tDecodeU64(&decoder, &pInfo->scanGroupId) < 0) return -1;
+  if (tDecodeU64(&decoder, &pInfo->maxVersion) < 0) return -1;
+
+  tEndDecode(&decoder);
+
+  tDecoderClear(&decoder);
+  return 0;
+}
--- a/source/libs/stream/test/tstreamUpdateTest.cpp
+++ b/source/libs/stream/test/tstreamUpdateTest.cpp
@ -6,11 +6,37 @@
 using namespace std;
 #define MAX_NUM_SCALABLE_BF    100000

+bool equalSBF(SScalableBf* left, SScalableBf* right) {
+  if (left->growth != right->growth) return false;
+  if (left->numBits != right->numBits) return false;
+  int lsize = taosArrayGetSize(left->bfArray);
+  int rsize = taosArrayGetSize(right->bfArray);
+  if (lsize != rsize) return false;
+  for (int32_t i = 0; i < lsize; i++) {
+    SBloomFilter* pLeftBF = (SBloomFilter*)taosArrayGetP(left->bfArray, i);
+    SBloomFilter* pRightBF = (SBloomFilter*)taosArrayGetP(right->bfArray, i);
+    if (pLeftBF->errorRate != pRightBF->errorRate) return false;
+    if (pLeftBF->expectedEntries != pRightBF->expectedEntries) return false;
+    if (pLeftBF->hashFn1 != pRightBF->hashFn1) return false;
+    if (pLeftBF->hashFn2 != pRightBF->hashFn2) return false;
+    if (pLeftBF->hashFunctions != pRightBF->hashFunctions) return false;
+    if (pLeftBF->numBits != pRightBF->numBits) return false;
+    if (pLeftBF->numUnits != pRightBF->numUnits) return false;
+    if (pLeftBF->size != pRightBF->size) return false;
+    uint64_t* leftUint = (uint64_t*) pLeftBF->buffer;
+    uint64_t* rightUint = (uint64_t*) pRightBF->buffer;
+    for (int32_t j = 0; j < pLeftBF->numUnits; j++) {
+      if (leftUint[j] != rightUint[j]) return false;
+    }
+  }
+  return true;
+}
+
 TEST(TD_STREAM_UPDATE_TEST, update) {
-  int64_t interval = 20 * 1000;
-  int64_t watermark = 10 * 60 * 1000;
+  const int64_t interval = 20 * 1000;
+  const int64_t watermark = 10 * 60 * 1000;
  SUpdateInfo *pSU = updateInfoInit(interval, TSDB_TIME_PRECISION_MILLI, watermark);
-  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,1, 0), true);
+  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,1, 0), false);
  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,1, -1), true);

  for(int i=0; i < 1024; i++) {
@ -31,15 +57,16 @@ TEST(TD_STREAM_UPDATE_TEST, update) {
    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,i, 1), true);
  }

+  TSKEY uid = 0;
  for(int i=3; i < 1024; i++) {
-    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,0, i), false);
+    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU, uid, i), false);
  }
-  GTEST_ASSERT_EQ(*(int64_t*)taosArrayGet(pSU->pTsBuckets,0), 1023);
+  GTEST_ASSERT_EQ(*(TSKEY*)taosHashGet(pSU->pMap, &uid, sizeof(uint64_t)), 1023);

  for(int i=3; i < 1024; i++) {
-    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU,0, i), true);
+    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU, uid, i), true);
  }
-  GTEST_ASSERT_EQ(*(int64_t*)taosArrayGet(pSU->pTsBuckets,0), 1023);
+  GTEST_ASSERT_EQ(*(TSKEY*)taosHashGet(pSU->pMap, &uid, sizeof(uint64_t)), 1023);

  SUpdateInfo *pSU1 = updateInfoInit(interval, TSDB_TIME_PRECISION_MILLI, watermark);
  for(int i=1; i <= watermark / interval; i++) {
@ -75,7 +102,8 @@ TEST(TD_STREAM_UPDATE_TEST, update) {
    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU2, 1, i * interval + 5), false);
    GTEST_ASSERT_EQ(pSU2->minTS, (i-(pSU2->numSBFs-1))*interval);
    GTEST_ASSERT_EQ(pSU2->numSBFs, watermark / interval);
-    GTEST_ASSERT_EQ(*(int64_t*)taosArrayGet(pSU2->pTsBuckets,1), i * interval + 5);
+    TSKEY uid2 = 1;
+    GTEST_ASSERT_EQ(*(TSKEY*)taosHashGet(pSU2->pMap, &uid2, sizeof(uint64_t)), i * interval + 5);
  }
  
  SUpdateInfo *pSU3 = updateInfoInit(interval, TSDB_TIME_PRECISION_MILLI, watermark);
@ -84,7 +112,8 @@ TEST(TD_STREAM_UPDATE_TEST, update) {
      GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU3, i, i * interval + 5 * j), false);
      GTEST_ASSERT_EQ(pSU3->minTS, 0);
      GTEST_ASSERT_EQ(pSU3->numSBFs, watermark / interval);
-      GTEST_ASSERT_EQ(*(int64_t*)taosArrayGet(pSU3->pTsBuckets, i), i * interval + 5 * j);
+      uint64_t uid3 = i;
+      GTEST_ASSERT_EQ(*(TSKEY*)taosHashGet(pSU3->pMap, &uid3, sizeof(uint64_t)), i * interval + 5 * j);
      SScalableBf *pSBF = (SScalableBf *)taosArrayGetP(pSU3->pTsSBFs, i);
      SBloomFilter *pBF = (SBloomFilter *)taosArrayGetP(pSBF->bfArray, 0);
      GTEST_ASSERT_EQ(pBF->size, j);
@ -92,13 +121,66 @@ TEST(TD_STREAM_UPDATE_TEST, update) {
  }

  SUpdateInfo *pSU4 = updateInfoInit(-1, TSDB_TIME_PRECISION_MILLI, -1);
-  GTEST_ASSERT_EQ(pSU4->watermark, MAX_NUM_SCALABLE_BF * pSU4->interval);
+  GTEST_ASSERT_EQ(pSU4->watermark, pSU4->interval);
  GTEST_ASSERT_EQ(pSU4->interval, MILLISECOND_PER_MINUTE);

  SUpdateInfo *pSU5 = updateInfoInit(0, TSDB_TIME_PRECISION_MILLI, 0);
-  GTEST_ASSERT_EQ(pSU5->watermark, MAX_NUM_SCALABLE_BF * pSU4->interval);
+  GTEST_ASSERT_EQ(pSU5->watermark, pSU4->interval);
  GTEST_ASSERT_EQ(pSU5->interval, MILLISECOND_PER_MINUTE);

+  SUpdateInfo *pSU7 = updateInfoInit(interval, TSDB_TIME_PRECISION_MILLI, watermark);
+  updateInfoAddCloseWindowSBF(pSU7);
+  for(int64_t i = 1; i < 2048000; i++) {
+    GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU7,i, i), false);
+  }
+  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU7,100, 1), true);
+  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU7,110, 10), true);
+  GTEST_ASSERT_EQ(updateInfoIsUpdated(pSU7,200, 20), true);
+
+  int32_t bufLen = updateInfoSerialize(NULL, 0, pSU7);
+  void* buf = taosMemoryCalloc(1, bufLen);
+  int32_t resSize = updateInfoSerialize(buf, bufLen, pSU7);
+
+  SUpdateInfo *pSU6 = updateInfoInit(0, TSDB_TIME_PRECISION_MILLI, 0);
+  int32_t desSize = updateInfoDeserialize(buf, bufLen, pSU6);
+  GTEST_ASSERT_EQ(desSize, 0);
+
+  GTEST_ASSERT_EQ(pSU7->interval, pSU6->interval);
+  GTEST_ASSERT_EQ(pSU7->maxVersion, pSU6->maxVersion);
+  GTEST_ASSERT_EQ(pSU7->minTS, pSU6->minTS);
+  GTEST_ASSERT_EQ(pSU7->numBuckets, pSU6->numBuckets);
+  GTEST_ASSERT_EQ(pSU7->numSBFs, pSU6->numSBFs);
+  GTEST_ASSERT_EQ(pSU7->scanGroupId,  pSU6->scanGroupId);
+  GTEST_ASSERT_EQ(pSU7->scanWindow.ekey, pSU6->scanWindow.ekey);
+  GTEST_ASSERT_EQ(pSU7->scanWindow.skey, pSU6->scanWindow.skey);
+  GTEST_ASSERT_EQ(pSU7->watermark, pSU6->watermark);
+  GTEST_ASSERT_EQ(equalSBF(pSU7->pCloseWinSBF, pSU6->pCloseWinSBF), true);
+
+  int32_t mapSize = taosHashGetSize(pSU7->pMap);
+  GTEST_ASSERT_EQ(mapSize, taosHashGetSize(pSU6->pMap));
+  void*  pIte = NULL;
+  size_t keyLen = 0;
+  while ((pIte = taosHashIterate(pSU7->pMap, pIte)) != NULL) {
+    void* key = taosHashGetKey(pIte, &keyLen);
+    void* value6 = taosHashGet(pSU6->pMap, key, keyLen);
+    GTEST_ASSERT_EQ(*(TSKEY*)pIte, *(TSKEY*)value6);
+  }
+
+  int32_t buSize = taosArrayGetSize(pSU7->pTsBuckets);
+  GTEST_ASSERT_EQ(buSize, taosArrayGetSize(pSU6->pTsBuckets));
+  for (int32_t i = 0; i < buSize; i++) {
+    TSKEY ts1 = *(TSKEY*)taosArrayGet(pSU7->pTsBuckets, i);
+    TSKEY ts2 = *(TSKEY*)taosArrayGet(pSU6->pTsBuckets, i);
+    GTEST_ASSERT_EQ(ts1, ts2);
+  }
+  int32_t lSize = taosArrayGetSize(pSU7->pTsSBFs);
+  int32_t rSize = taosArrayGetSize(pSU6->pTsSBFs);
+  GTEST_ASSERT_EQ(lSize, rSize);
+  for (int32_t i = 0; i < lSize; i++) {
+    SScalableBf* pLeftSBF = (SScalableBf*)taosArrayGetP(pSU7->pTsSBFs, i);
+    SScalableBf* pRightSBF = (SScalableBf*)taosArrayGetP(pSU6->pTsSBFs, i);
+    GTEST_ASSERT_EQ(equalSBF(pLeftSBF, pRightSBF), true);
+  }

  updateInfoDestroy(pSU);
  updateInfoDestroy(pSU1);
@ -106,6 +188,9 @@ TEST(TD_STREAM_UPDATE_TEST, update) {
  updateInfoDestroy(pSU3);
  updateInfoDestroy(pSU4);
  updateInfoDestroy(pSU5);
+  updateInfoDestroy(pSU6);
+  updateInfoDestroy(pSU7);
+
 }

 int main(int argc, char* argv[]) {
--- a/source/libs/sync/inc/syncInt.h
+++ b/source/libs/sync/inc/syncInt.h
@ -162,6 +162,9 @@ typedef struct SSyncNode {
  // is config changing
  bool changing;

+  int64_t startTime;
+  int64_t lastReplicateTime;
+
 } SSyncNode;

 // open/close --------------
@ -186,16 +189,18 @@ int32_t syncNodePingAll(SSyncNode* pSyncNode);
 // timer control --------------
 int32_t syncNodeStartPingTimer(SSyncNode* pSyncNode);
 int32_t syncNodeStopPingTimer(SSyncNode* pSyncNode);
+
 int32_t syncNodeStartElectTimer(SSyncNode* pSyncNode, int32_t ms);
 int32_t syncNodeStopElectTimer(SSyncNode* pSyncNode);
 int32_t syncNodeRestartElectTimer(SSyncNode* pSyncNode, int32_t ms);
 int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode);
+
 int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode);
-int32_t syncNodeStartNowHeartbeatTimer(SSyncNode* pSyncNode);
+int32_t syncNodeStartHeartbeatTimerNow(SSyncNode* pSyncNode);
 int32_t syncNodeStartHeartbeatTimerMS(SSyncNode* pSyncNode, int32_t ms);
 int32_t syncNodeStopHeartbeatTimer(SSyncNode* pSyncNode);
 int32_t syncNodeRestartHeartbeatTimer(SSyncNode* pSyncNode);
-int32_t syncNodeRestartNowHeartbeatTimer(SSyncNode* pSyncNode);
+int32_t syncNodeRestartHeartbeatTimerNow(SSyncNode* pSyncNode);
 int32_t syncNodeRestartNowHeartbeatTimerMS(SSyncNode* pSyncNode, int32_t ms);

 // utils --------------
--- a/source/libs/sync/inc/syncReplication.h
+++ b/source/libs/sync/inc/syncReplication.h
@ -55,7 +55,7 @@ int32_t syncNodeAppendEntriesPeers(SSyncNode* pSyncNode);
 int32_t syncNodeAppendEntriesPeersSnapshot(SSyncNode* pSyncNode);
 int32_t syncNodeAppendEntriesPeersSnapshot2(SSyncNode* pSyncNode);

-int32_t syncNodeReplicate(SSyncNode* pSyncNode);
+int32_t syncNodeReplicate(SSyncNode* pSyncNode, bool isTimer);
 int32_t syncNodeAppendEntries(SSyncNode* pSyncNode, const SRaftId* destRaftId, const SyncAppendEntries* pMsg);
 int32_t syncNodeAppendEntriesBatch(SSyncNode* pSyncNode, const SRaftId* destRaftId, const SyncAppendEntriesBatch* pMsg);

--- a/source/libs/sync/src/syncMain.c
+++ b/source/libs/sync/src/syncMain.c
@ -495,6 +495,18 @@ const char* syncGetMyRoleStr(int64_t rid) {
  return s;
 }

+bool syncRestoreFinish(int64_t rid) {
+  SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
+  if (pSyncNode == NULL) {
+    return false;
+  }
+  ASSERT(rid == pSyncNode->rid);
+  bool restoreFinish = pSyncNode->restoreFinish;
+
+  taosReleaseRef(tsNodeRefId, pSyncNode->rid);
+  return restoreFinish;
+}
+
 SyncTerm syncGetMyTerm(int64_t rid) {
  SSyncNode* pSyncNode = (SSyncNode*)taosAcquireRef(tsNodeRefId, rid);
  if (pSyncNode == NULL) {
@ -1086,6 +1098,10 @@ SSyncNode* syncNodeOpen(const SSyncInfo* pOldSyncInfo) {
  // start raft
  // syncNodeBecomeFollower(pSyncNode);

+  int64_t timeNow = taosGetTimestampMs();
+  pSyncNode->startTime = timeNow;
+  pSyncNode->lastReplicateTime = timeNow;
+
  syncNodeEventLog(pSyncNode, "sync open");

  return pSyncNode;
@ -1303,7 +1319,7 @@ int32_t syncNodeResetElectTimer(SSyncNode* pSyncNode) {
  return ret;
 }

-int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode) {
+static int32_t syncNodeDoStartHeartbeatTimer(SSyncNode* pSyncNode) {
  int32_t ret = 0;
  if (syncEnvIsStart()) {
    taosTmrReset(pSyncNode->FpHeartbeatTimerCB, pSyncNode->heartbeatTimerMS, pSyncNode, gSyncEnv->pTimerManager,
@ -1322,26 +1338,21 @@ int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode) {
  return ret;
 }

-int32_t syncNodeStartHeartbeatTimerMS(SSyncNode* pSyncNode, int32_t ms) {
-  int32_t ret = 0;
-  if (syncEnvIsStart()) {
-    taosTmrReset(pSyncNode->FpHeartbeatTimerCB, ms, pSyncNode, gSyncEnv->pTimerManager, &pSyncNode->pHeartbeatTimer);
-    atomic_store_64(&pSyncNode->heartbeatTimerLogicClock, pSyncNode->heartbeatTimerLogicClockUser);
-  } else {
-    sError("vgId:%d, start heartbeat timer error, sync env is stop", pSyncNode->vgId);
-  }
-
-  do {
-    char logBuf[128];
-    snprintf(logBuf, sizeof(logBuf), "start heartbeat timer, ms:%d", ms);
-    syncNodeEventLog(pSyncNode, logBuf);
-  } while (0);
-
+int32_t syncNodeStartHeartbeatTimer(SSyncNode* pSyncNode) {
+  pSyncNode->heartbeatTimerMS = pSyncNode->hbBaseLine;
+  int32_t ret = syncNodeDoStartHeartbeatTimer(pSyncNode);
  return ret;
 }

-int32_t syncNodeStartNowHeartbeatTimer(SSyncNode* pSyncNode) {
-  int32_t ret = syncNodeStartHeartbeatTimerMS(pSyncNode, 1);
+int32_t syncNodeStartHeartbeatTimerMS(SSyncNode* pSyncNode, int32_t ms) {
+  pSyncNode->heartbeatTimerMS = ms;
+  int32_t ret = syncNodeDoStartHeartbeatTimer(pSyncNode);
+  return ret;
+}
+
+int32_t syncNodeStartHeartbeatTimerNow(SSyncNode* pSyncNode) {
+  pSyncNode->heartbeatTimerMS = 1;
+  int32_t ret = syncNodeDoStartHeartbeatTimer(pSyncNode);
  return ret;
 }

@ -1362,9 +1373,9 @@ int32_t syncNodeRestartHeartbeatTimer(SSyncNode* pSyncNode) {
  return 0;
 }

-int32_t syncNodeRestartNowHeartbeatTimer(SSyncNode* pSyncNode) {
+int32_t syncNodeRestartHeartbeatTimerNow(SSyncNode* pSyncNode) {
  syncNodeStopHeartbeatTimer(pSyncNode);
-  syncNodeStartNowHeartbeatTimer(pSyncNode);
+  syncNodeStartHeartbeatTimerNow(pSyncNode);
  return 0;
 }

@ -1942,9 +1953,6 @@ void syncNodeDoConfigChange(SSyncNode* pSyncNode, SSyncCfg* pNewConfig, SyncInde

      // Raft 3.6.2 Committing entries from previous terms
      syncNodeAppendNoop(pSyncNode);
-#if 0  // simon
-      syncNodeReplicate(pSyncNode);
-#endif
      syncMaybeAdvanceCommitIndex(pSyncNode);

    } else {
@ -2126,9 +2134,6 @@ void syncNodeCandidate2Leader(SSyncNode* pSyncNode) {

  // Raft 3.6.2 Committing entries from previous terms
  syncNodeAppendNoop(pSyncNode);
-#if 0  // simon
-  syncNodeReplicate(pSyncNode);
-#endif
  syncMaybeAdvanceCommitIndex(pSyncNode);
 }

@ -2499,7 +2504,7 @@ static int32_t syncNodeAppendNoop(SSyncNode* ths) {
  if (ths->state == TAOS_SYNC_STATE_LEADER) {
    int32_t code = ths->pLogStore->syncLogAppendEntry(ths->pLogStore, pEntry);
    ASSERT(code == 0);
-    syncNodeReplicate(ths);
+    syncNodeReplicate(ths, false);
  }

  syncEntryDestory(pEntry);
@ -2572,7 +2577,7 @@ int32_t syncNodeOnClientRequestCb(SSyncNode* ths, SyncClientRequest* pMsg, SyncI

    // if mulit replica, start replicate right now
    if (ths->replicaNum > 1) {
-      syncNodeReplicate(ths);
+      syncNodeReplicate(ths, false);

      // pre commit
      syncNodePreCommit(ths, pEntry, 0);
@ -2641,7 +2646,7 @@ int32_t syncNodeOnClientRequestBatchCb(SSyncNode* ths, SyncClientRequestBatch* p

  if (ths->replicaNum > 1) {
    // if multi replica, start replicate right now
-    syncNodeReplicate(ths);
+    syncNodeReplicate(ths, false);

  } else if (ths->replicaNum == 1) {
    // one replica
--- a/source/libs/sync/src/syncReplication.c
+++ b/source/libs/sync/src/syncReplication.c
@ -202,17 +202,19 @@ int32_t syncNodeAppendEntriesPeersSnapshot2(SSyncNode* pSyncNode) {
    syncAppendEntriesBatchDestroy(pMsg);

    // speed up
-    if (pMsg->dataCount > 0 && pMsg->prevLogIndex < pSyncNode->commitIndex) {
+    if (pMsg->dataCount > 0 && pSyncNode->commitIndex - pMsg->prevLogIndex > SYNC_SLOW_DOWN_RANGE) {
      ret = 1;

+#if 0
      do {
        char     logBuf[128];
        char     host[64];
        uint16_t port;
        syncUtilU642Addr(pDestId->addr, host, sizeof(host), &port);
-        snprintf(logBuf, sizeof(logBuf), "speed up for %s:%d, pre-index:%ld", host, port, pMsg->prevLogIndex);
+        snprintf(logBuf, sizeof(logBuf), "maybe speed up for %s:%d, pre-index:%ld", host, port, pMsg->prevLogIndex);
        syncNodeEventLog(pSyncNode, logBuf);
      } while (0);
+#endif
    }
  }

@ -301,7 +303,7 @@ int32_t syncNodeAppendEntriesPeersSnapshot(SSyncNode* pSyncNode) {
  return ret;
 }

-int32_t syncNodeReplicate(SSyncNode* pSyncNode) {
+int32_t syncNodeReplicate(SSyncNode* pSyncNode, bool isTimer) {
  // start replicate
  int32_t ret = 0;

@ -323,13 +325,38 @@ int32_t syncNodeReplicate(SSyncNode* pSyncNode) {
      break;
  }

-  if (ret > 0) {
+  // start delay
+  int64_t timeNow = taosGetTimestampMs();
+  int64_t startDelay = timeNow - pSyncNode->startTime;
+
+  // replicate delay
+  int64_t replicateDelay = timeNow - pSyncNode->lastReplicateTime;
+  pSyncNode->lastReplicateTime = timeNow;
+
+  if (ret > 0 && isTimer && startDelay > SYNC_SPEED_UP_AFTER_MS) {
    // speed up replicate
-    int32_t ms = pSyncNode->heartbeatTimerMS < 50 ? pSyncNode->heartbeatTimerMS : 50;
+    int32_t ms =
+        pSyncNode->heartbeatTimerMS < SYNC_SPEED_UP_HB_TIMER ? pSyncNode->heartbeatTimerMS : SYNC_SPEED_UP_HB_TIMER;
    syncNodeRestartNowHeartbeatTimerMS(pSyncNode, ms);

+#if 0
+    do {
+      char logBuf[128];
+      snprintf(logBuf, sizeof(logBuf), "replicate speed up");
+      syncNodeEventLog(pSyncNode, logBuf);
+    } while (0);
+#endif
+
  } else {
    syncNodeRestartHeartbeatTimer(pSyncNode);
+
+#if 0
+    do {
+      char logBuf[128];
+      snprintf(logBuf, sizeof(logBuf), "replicate slow down");
+      syncNodeEventLog(pSyncNode, logBuf);
+    } while (0);
+#endif
  }

  return ret;
--- a/source/libs/sync/src/syncTimeout.c
+++ b/source/libs/sync/src/syncTimeout.c
@ -58,7 +58,7 @@ int32_t syncNodeOnTimeoutCb(SSyncNode* ths, SyncTimeout* pMsg) {
      ++(ths->heartbeatTimerCounter);
      sInfo("vgId:%d, sync timeout, type:replicate count:%d, heartbeatTimerLogicClockUser:%ld", ths->vgId,
            ths->heartbeatTimerCounter, ths->heartbeatTimerLogicClockUser);
-      syncNodeReplicate(ths);
+      syncNodeReplicate(ths, true);
    }
  } else {
    sError("vgId:%d, unknown timeout-type:%d", ths->vgId, pMsg->timeoutType);
--- a/source/libs/transport/src/transCli.c
+++ b/source/libs/transport/src/transCli.c
@ -1271,6 +1271,9 @@ int cliAppCb(SCliConn* pConn, STransMsg* pResp, SCliMsg* pMsg) {
        transFreeMsg(pResp->pCont);
        cliSchedMsgToNextNode(pMsg, pThrd);
        return -1;
+      } else {
+        // change error code for taos client driver if retryCnt exceeds limit
+        if (0 == strncmp(pTransInst->label, "TSC", strlen("TSC"))) pResp->code = TSDB_CODE_APP_NOT_READY;
      }
    }
  }
@ -1381,6 +1384,7 @@ int transReleaseCliHandle(void* handle) {
  tGDebug("send release request at thread:%08" PRId64 "", pThrd->pid);

  if (0 != transAsyncSend(pThrd->asyncPool, &cmsg->q)) {
+    taosMemoryFree(cmsg);
    return -1;
  }
  return 0;
--- a/source/os/src/osFile.c
+++ b/source/os/src/osFile.c
@ -703,7 +703,11 @@ int64_t taosFSendFile(TdFilePtr pFileOut, TdFilePtr pFileIn, int64_t *offset, in
  int64_t sentbytes;

  while (leftbytes > 0) {
+  #ifdef _TD_ARM_32
+    sentbytes = sendfile(pFileOut->fd, pFileIn->fd, (long int*)offset, leftbytes);
+  #else
    sentbytes = sendfile(pFileOut->fd, pFileIn->fd, offset, leftbytes);
+  #endif
    if (sentbytes == -1) {
      if (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK) {
        continue;
--- a/source/util/src/tbloomfilter.c
+++ b/source/util/src/tbloomfilter.c
@ -108,8 +108,41 @@ void tBloomFilterDestroy(SBloomFilter *pBF) {
  taosMemoryFree(pBF);
 }

-void tBloomFilterDump(const struct SBloomFilter *pBF) {
-// ToDo
+int32_t tBloomFilterEncode(const SBloomFilter *pBF, SEncoder* pEncoder) {
+  if (tEncodeU32(pEncoder, pBF->hashFunctions) < 0) return -1;
+  if (tEncodeU64(pEncoder, pBF->expectedEntries) < 0) return -1;
+  if (tEncodeU64(pEncoder, pBF->numUnits) < 0) return -1;
+  if (tEncodeU64(pEncoder, pBF->numBits) < 0) return -1;
+  if (tEncodeU64(pEncoder, pBF->size) < 0) return -1;
+  for (uint64_t i = 0; i < pBF->numUnits; i++) {
+    uint64_t* pUnits = (uint64_t*)pBF->buffer;
+    if (tEncodeU64(pEncoder, pUnits[i]) < 0) return -1;
+  }
+  if (tEncodeDouble(pEncoder, pBF->errorRate) < 0) return -1;
+  return 0;
+}
+
+SBloomFilter* tBloomFilterDecode(SDecoder* pDecoder) {
+  SBloomFilter *pBF = taosMemoryCalloc(1, sizeof(SBloomFilter));
+  pBF->buffer = NULL;
+  if (tDecodeU32(pDecoder, &pBF->hashFunctions) < 0) goto _error;
+  if (tDecodeU64(pDecoder, &pBF->expectedEntries) < 0) goto _error;
+  if (tDecodeU64(pDecoder, &pBF->numUnits) < 0) goto _error;
+  if (tDecodeU64(pDecoder, &pBF->numBits) < 0) goto _error;
+  if (tDecodeU64(pDecoder, &pBF->size) < 0) goto _error;
+  pBF->buffer = taosMemoryCalloc(pBF->numUnits, sizeof(uint64_t));
+  for (int32_t i = 0; i < pBF->numUnits; i++) {
+    uint64_t* pUnits = (uint64_t*)pBF->buffer;
+    if (tDecodeU64(pDecoder, pUnits + i) < 0) goto _error;
+  }
+  if (tDecodeDouble(pDecoder, &pBF->errorRate) < 0) goto _error;
+  pBF->hashFn1 = taosGetDefaultHashFunction(TSDB_DATA_TYPE_TIMESTAMP);
+  pBF->hashFn2 = taosGetDefaultHashFunction(TSDB_DATA_TYPE_NCHAR);
+  return pBF;
+
+_error:
+  tBloomFilterDestroy(pBF);
+  return NULL;
 }

 bool tBloomFilterIsFull(const SBloomFilter *pBF) {
--- a/source/util/src/tcache.c
+++ b/source/util/src/tcache.c
@ -35,7 +35,7 @@ typedef struct SCacheNode {
  uint64_t           addedTime;   // the added time when this element is added or updated into cache
  uint64_t           lifespan;    // life duration when this element should be remove from cache
  int64_t            expireTime;  // expire time
-  uint64_t           signature;
+  void*              signature;
  struct STrashElem *pTNodeHeader;    // point to trash node head
  uint16_t           keyLen : 15;     // max key size: 32kb
  bool               inTrashcan : 1;  // denote if it is in trash or not
@ -208,7 +208,7 @@ static void taosTrashcanEmpty(SCacheObj *pCacheObj, bool force);
 * @param pNode          data node
 */
 static FORCE_INLINE void taosCacheReleaseNode(SCacheObj *pCacheObj, SCacheNode *pNode) {
-  if (pNode->signature != (uint64_t)pNode) {
+  if (pNode->signature != pNode) {
    uError("key:%s, %p data is invalid, or has been released", pNode->key, pNode);
    return;
  }
@ -226,7 +226,7 @@ static FORCE_INLINE void taosCacheReleaseNode(SCacheObj *pCacheObj, SCacheNode *
 }

 static FORCE_INLINE STrashElem *doRemoveElemInTrashcan(SCacheObj *pCacheObj, STrashElem *pElem) {
-  if (pElem->pData->signature != (uint64_t)pElem->pData) {
+  if (pElem->pData->signature != pElem->pData) {
    uWarn("key:sig:0x%" PRIx64 " %p data has been released, ignore", pElem->pData->signature, pElem->pData);
    return NULL;
  }
@ -494,7 +494,7 @@ void *taosCacheAcquireByData(SCacheObj *pCacheObj, void *data) {
  if (pCacheObj == NULL || data == NULL) return NULL;

  SCacheNode *ptNode = (SCacheNode *)((char *)data - sizeof(SCacheNode));
-  if (ptNode->signature != (uint64_t)ptNode) {
+  if (ptNode->signature != ptNode) {
    uError("cache:%s, key: %p the data from cache is invalid", pCacheObj->name, ptNode);
    return NULL;
  }
@ -511,7 +511,7 @@ void *taosCacheTransferData(SCacheObj *pCacheObj, void **data) {
  if (pCacheObj == NULL || data == NULL || (*data) == NULL) return NULL;

  SCacheNode *ptNode = (SCacheNode *)((char *)(*data) - sizeof(SCacheNode));
-  if (ptNode->signature != (uint64_t)ptNode) {
+  if (ptNode->signature != ptNode) {
    uError("cache:%s, key: %p the data from cache is invalid", pCacheObj->name, ptNode);
    return NULL;
  }
@ -539,7 +539,7 @@ void taosCacheRelease(SCacheObj *pCacheObj, void **data, bool _remove) {
  // It happens when there is only one object in the cache, and two threads which has referenced this object
  // start to free the it simultaneously [TD-1569].
  SCacheNode *pNode = (SCacheNode *)((char *)(*data) - sizeof(SCacheNode));
-  if (pNode->signature != (uint64_t)pNode) {
+  if (pNode->signature != pNode) {
    uError("cache:%s, %p, release invalid cache data", pCacheObj->name, pNode);
    return;
  }
@ -728,7 +728,7 @@ SCacheNode *taosCreateCacheNode(const char *key, size_t keyLen, const char *pDat
  pNewNode->addedTime = (uint64_t)taosGetTimestampMs();
  pNewNode->lifespan = duration;
  pNewNode->expireTime = pNewNode->addedTime + pNewNode->lifespan;
-  pNewNode->signature = (uint64_t)pNewNode;
+  pNewNode->signature = pNewNode;
  pNewNode->size = (uint32_t)sizeInBytes;

  return pNewNode;
--- a/source/util/src/terror.c
+++ b/source/util/src/terror.c
@ -47,7 +47,7 @@ STaosError errors[] = {

 // rpc
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_AUTH_FAILURE,             "Authentication failure")
-TAOS_DEFINE_ERROR(TSDB_CODE_RPC_REDIRECT,                 "Redirect")
+TAOS_DEFINE_ERROR(TSDB_CODE_RPC_REDIRECT,                 "Database not ready, need retry")
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_NETWORK_UNAVAIL,          "Unable to establish connection")
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_FQDN_ERROR,               "Unable to resolve FQDN")
 TAOS_DEFINE_ERROR(TSDB_CODE_RPC_PORT_EADDRINUSE,          "Port already in use")
--- a/source/util/src/tscalablebf.c
+++ b/source/util/src/tscalablebf.c
@ -101,6 +101,42 @@ void tScalableBfDestroy(SScalableBf *pSBf) {
  taosMemoryFree(pSBf);
 }

-void tScalableBfDump(const SScalableBf *pSBf) {
-  // Todo;
+int32_t tScalableBfEncode(const SScalableBf *pSBf, SEncoder* pEncoder) {
+  if (!pSBf) {
+    if (tEncodeI32(pEncoder, 0) < 0) return -1;
+    return 0;
+  }
+  int32_t size = taosArrayGetSize(pSBf->bfArray);
+  if (tEncodeI32(pEncoder, size) < 0) return -1;
+  for (int32_t i = 0; i < size; i++) {
+    SBloomFilter* pBF = taosArrayGetP(pSBf->bfArray, i);
+    if (tBloomFilterEncode(pBF, pEncoder) < 0) return -1;
+  }
+  if (tEncodeU32(pEncoder, pSBf->growth) < 0) return -1;
+  if (tEncodeU64(pEncoder, pSBf->numBits) < 0) return -1;
+  return 0;
+}
+
+SScalableBf* tScalableBfDecode(SDecoder* pDecoder) {
+  SScalableBf *pSBf = taosMemoryCalloc(1, sizeof(SScalableBf));
+  pSBf->bfArray = NULL;
+  int32_t size = 0;
+  if (tDecodeI32(pDecoder, &size) < 0) goto _error;
+  if (size == 0) {
+    tScalableBfDestroy(pSBf);
+    return NULL;
+  }
+  pSBf->bfArray = taosArrayInit(size * 2, sizeof(void *));
+  for (int32_t i = 0; i < size; i++) {
+    SBloomFilter* pBF = tBloomFilterDecode(pDecoder);
+    if (!pBF) goto _error;
+    taosArrayPush(pSBf->bfArray, &pBF);
+  }
+  if (tDecodeU32(pDecoder, &pSBf->growth) < 0) goto _error;
+  if (tDecodeU64(pDecoder, &pSBf->numBits) < 0) goto _error;
+  return pSBf;
+
+_error:
+  tScalableBfDestroy(pSBf);
+  return NULL;
 }
--- a/tests/pytest/util/dnodes-random-fail.py
+++ b/tests/pytest/util/dnodes-random-fail.py
@ -24,12 +24,9 @@ class TDSimClient:

        self.cfgDict = {
            "numOfLogLines": "100000000",
-            "numOfThreadsPerCore": "2.0",
            "locale": "en_US.UTF-8",
            "charset": "UTF-8",
            "asyncLog": "0",
-            "anyIp": "0",
-            "sdbDebugFlag": "135",
            "rpcDebugFlag": "135",
            "tmrDebugFlag": "131",
            "cDebugFlag": "135",
--- a/tests/pytest/util/dnodes.py
+++ b/tests/pytest/util/dnodes.py
@ -30,7 +30,6 @@ class TDSimClient:
        self.path = path
        self.cfgDict = {
            "numOfLogLines": "100000000",
-            "numOfThreadsPerCore": "2.0",
            "locale": "en_US.UTF-8",
            "charset": "UTF-8",
            "asyncLog": "0",
@ -40,6 +39,7 @@ class TDSimClient:
            "udebugFlag": "143",
            "jnidebugFlag": "143",
            "qdebugFlag": "143",
+            "supportVnodes": "1024",
            "telemetryReporting": "0",
        }

@ -117,8 +117,6 @@ class TDDnode:
        self.valgrind = 0
        self.remoteIP = ""
        self.cfgDict = {
-            "walLevel": "2",
-            "fsync": "1000",
            "monitor": "0",
            "maxShellConns": "30000",
            "locale": "en_US.UTF-8",
@ -139,6 +137,7 @@ class TDDnode:
            "qdebugFlag": "143",
            "numOfLogLines": "100000000",
            "statusInterval": "1",
+            "supportVnodes": "1024",
            "telemetryReporting": "0"
        }

--- a/tests/script/jenkins/basic.txt
+++ b/tests/script/jenkins/basic.txt
@ -84,7 +84,7 @@
 ./test.sh -f tsim/insert/update0.sim
 ./test.sh -f tsim/insert/update1_sort_merge.sim

-# ---- parser
+# ---- parser ----
 ./test.sh -f tsim/parser/alter__for_community_version.sim
 ./test.sh -f tsim/parser/alter_column.sim
 ./test.sh -f tsim/parser/alter_stable.sim
@ -106,7 +106,7 @@
 ./test.sh -f tsim/parser/commit.sim
 ./test.sh -f tsim/parser/condition.sim
 ./test.sh -f tsim/parser/constCol.sim
-#./test.sh -f tsim/parser/create_db.sim
+./test.sh -f tsim/parser/create_db.sim
 ./test.sh -f tsim/parser/create_mt.sim
 ./test.sh -f tsim/parser/create_tb_with_tag_name.sim
 ./test.sh -f tsim/parser/create_tb.sim
@ -129,7 +129,7 @@
 ./test.sh -f tsim/parser/import.sim
 ./test.sh -f tsim/parser/insert_multiTbl.sim
 ./test.sh -f tsim/parser/insert_tb.sim
-# TD-17038 ./test.sh -f tsim/parser/interp.sim
+# TD-18293 ./test.sh -f tsim/parser/interp.sim
 ./test.sh -f tsim/parser/join_manyblocks.sim
 ./test.sh -f tsim/parser/join_multitables.sim
 ./test.sh -f tsim/parser/join_multivnode.sim
@ -318,7 +318,7 @@
 ./test.sh -f tsim/valgrind/checkError7.sim
 ./test.sh -f tsim/valgrind/checkUdf.sim

-# --- vnode
+# --- vnode ----
 # unsupport ./test.sh -f tsim/vnode/replica3_basic.sim
 # unsupport ./test.sh -f tsim/vnode/replica3_repeat.sim
 # unsupport ./test.sh -f tsim/vnode/replica3_vgroup.sim
--- a/tests/script/sh/compile_udf.sh
+++ b/tests/script/sh/compile_udf.sh
@ -4,7 +4,7 @@ rm -rf /tmp/udf/libbitand.so /tmp/udf/libsqrsum.so
 mkdir -p /tmp/udf
 echo "compile udf bit_and and sqr_sum"
 gcc -fPIC -shared sh/bit_and.c -I../../include/libs/function/ -I../../include/client -I../../include/util   -o /tmp/udf/libbitand.so
-gcc -fPIC -shared sh/sqr_sum.c -I../../include/libs/function/ -I../../include/client -I../../include/util   -o /tmp/udf/libsqrsum.so
+gcc -fPIC -shared sh/l2norm.c -I../../include/libs/function/ -I../../include/client -I../../include/util   -o /tmp/udf/libl2norm.so
 echo "debug show /tmp/udf/*.so"
 ls /tmp/udf/*.so

--- a/tests/script/sh/sqr_sum.c
+++ b/tests/script/sh/sqr_sum.c
@ -5,22 +5,22 @@

 #include "taosudf.h"

-DLL_EXPORT int32_t sqr_sum_init() {
+DLL_EXPORT int32_t l2norm_init() {
  return 0;
 }

-DLL_EXPORT int32_t sqr_sum_destroy() {
+DLL_EXPORT int32_t l2norm_destroy() {
  return 0;
 }

-DLL_EXPORT int32_t sqr_sum_start(SUdfInterBuf *buf) {
+DLL_EXPORT int32_t l2norm_start(SUdfInterBuf *buf) {
  *(int64_t*)(buf->buf) = 0;
  buf->bufLen = sizeof(double);
  buf->numOfResult = 0;
  return 0;
 }

-DLL_EXPORT int32_t sqr_sum(SUdfDataBlock* block, SUdfInterBuf *interBuf, SUdfInterBuf *newInterBuf) {
+DLL_EXPORT int32_t l2norm(SUdfDataBlock* block, SUdfInterBuf *interBuf, SUdfInterBuf *newInterBuf) {
  double sumSquares = *(double*)interBuf->buf;
  int8_t numNotNull = 0;
  for (int32_t i = 0; i < block->numOfCols; ++i) {
@ -67,7 +67,7 @@ DLL_EXPORT int32_t sqr_sum(SUdfDataBlock* block, SUdfInterBuf *interBuf, SUdfInt
  return 0;
 }

-DLL_EXPORT int32_t sqr_sum_finish(SUdfInterBuf* buf, SUdfInterBuf *resultData) {
+DLL_EXPORT int32_t l2norm_finish(SUdfInterBuf* buf, SUdfInterBuf *resultData) {
  if (buf->numOfResult == 0) {
    resultData->numOfResult = 0;
    return 0;
--- a/tests/script/tsim/db/alter_option.sim
+++ b/tests/script/tsim/db/alter_option.sim
@ -111,16 +111,16 @@ endi
 if $data21_db != 1000 then  # wal_level fsyncperiod
  return -1
 endi
-if $data22_db != 0 then  # 
+if $data22_db != 172800 then  # wal_retention_period
  return -1
 endi
-if $data23_db != 0 then  # 
+if $data23_db != -1 then  # wal_retention_size
  return -1
 endi
-if $data24_db != 0 then  # 
+if $data24_db != 86400 then  # wal_roll_period
  return -1
 endi
-if $data25_db != 0 then  # 
+if $data25_db != 0 then  # wal_segment_size
  return -1
 endi

--- a/tests/script/tsim/parser/alter1.sim
+++ b/tests/script/tsim/parser/alter1.sim
@ -129,3 +129,5 @@ endi
 #if $rows != 0 then
 #  return -1
 #endi
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_bigint.sim
+++ b/tests/script/tsim/parser/columnValue_bigint.sim
@ -425,3 +425,5 @@ sql insert into st_bigint_e25 using mt_bigint tags (033) values (now, 00062)
 #sql_error alter table st_bigint_e23 set tag tagname="abc"
 #sql_error alter table st_bigint_e24 set tag tagname=" "
 #sql_error alter table st_bigint_e25 set tag tagname=''
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_bool.sim
+++ b/tests/script/tsim/parser/columnValue_bool.sim
@ -634,3 +634,4 @@ sql alter table st_bool_i4 set tag tagname="abc"
 sql alter table st_bool_i5 set tag tagname=" "     
 sql alter table st_bool_i6 set tag tagname=''      

+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_double.sim
+++ b/tests/script/tsim/parser/columnValue_double.sim
@ -528,3 +528,5 @@ sql_error alter table st_double_e22 set tag tagname=abc
 sql alter table st_double_e23 set tag tagname="abc"   
 sql alter table st_double_e24 set tag tagname=" "     
 sql alter table st_double_e25 set tag tagname=''      
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_float.sim
+++ b/tests/script/tsim/parser/columnValue_float.sim
@ -558,3 +558,5 @@ sql_error alter table st_float_e22 set tag tagname=abc
 sql alter table st_float_e23 set tag tagname="abc"   
 sql alter table st_float_e24 set tag tagname=" "     
 sql alter table st_float_e25 set tag tagname=''      
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_int.sim
+++ b/tests/script/tsim/parser/columnValue_int.sim
@ -423,3 +423,5 @@ sql_error alter table st_int_e22 set tag tagname=abc
 sql_error alter table st_int_e23 set tag tagname="abc"   
 sql_error alter table st_int_e24 set tag tagname=" "     
 sql alter table st_int_e25 set tag tagname=''      
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_smallint.sim
+++ b/tests/script/tsim/parser/columnValue_smallint.sim
@ -426,3 +426,5 @@ sql_error alter table st_smallint_e22 set tag tagname=abc
 sql_error alter table st_smallint_e23 set tag tagname="abc"   
 sql_error alter table st_smallint_e24 set tag tagname=" "     
 sql alter table st_smallint_e25 set tag tagname=''      
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_tinyint.sim
+++ b/tests/script/tsim/parser/columnValue_tinyint.sim
@ -424,3 +424,5 @@ sql_error alter table st_tinyint_e22 set tag tagname=abc
 sql_error alter table st_tinyint_e23 set tag tagname="abc"   
 sql_error alter table st_tinyint_e24 set tag tagname=" "     
 sql alter table st_tinyint_e25 set tag tagname=''      
+
+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/columnValue_unsign.sim
+++ b/tests/script/tsim/parser/columnValue_unsign.sim
@ -130,3 +130,4 @@ if $rows != 1 then
  return -1
 endi

+system sh/exec.sh -n dnode1 -s stop  -x SIGINT
--- a/tests/script/tsim/parser/create_db.sim
+++ b/tests/script/tsim/parser/create_db.sim
@ -237,8 +237,8 @@ if $rows != 3 then
  return -1
 endi
 sql show databases
-print wallevel $data13_testwal
-if $data13_testwal != 1 then 
+print wallevel $data20_testwal
+if $data20_testwal != 1 then 
  return -1
 endi
 sql drop database testwal
--- a/tests/script/tsim/query/udf.sim
+++ b/tests/script/tsim/query/udf.sim
@ -24,10 +24,10 @@ if $system_content == Windows_NT then
 endi
 if $system_content == Windows_NT then
  sql create function bit_and as 'C:\\Windows\\Temp\\bitand.dll' outputtype int bufSize 8;
-  sql create aggregate function sqr_sum as 'C:\\Windows\\Temp\\sqrsum.dll' outputtype double bufSize 8;
+  sql create aggregate function l2norm as 'C:\\Windows\\Temp\\l2norm.dll' outputtype double bufSize 8;
 else
  sql create function bit_and as '/tmp/udf/libbitand.so' outputtype int bufSize 8;
-  sql create aggregate function sqr_sum as '/tmp/udf/libsqrsum.so' outputtype double bufSize 8;
+  sql create aggregate function l2norm as '/tmp/udf/libl2norm.so' outputtype double bufSize 8;
 endi
 sql show functions;
 if $rows != 2 then
@ -44,7 +44,7 @@ if $data10 != 2 then
  return -1
 endi

-sql select sqr_sum(f) from t;
+sql select l2norm(f) from t;
 if $rows != 1 then
  print expect 1, actual $rows
  return -1
@ -66,7 +66,7 @@ if $data10 != 1 then
  return -1
 endi

-sql select sqr_sum(f1, f2) from t2;
+sql select l2norm(f1, f2) from t2;
 if $rows != 1 then
  return -1
 endi
@ -95,7 +95,7 @@ if $data30 != NULL then
  return -1
 endi

-sql select sqr_sum(f1, f2) from t2;
+sql select l2norm(f1, f2) from t2;
 print $rows, $data00
 if $rows != 1 then
  return -1
@ -105,7 +105,7 @@ if $data00 != 2.645751311 then
 endi

 sql insert into t2 values(now+4s, 4, 8)(now+5s, 5, 9);
-sql select sqr_sum(f1-f2), sqr_sum(f1+f2) from t2;
+sql select l2norm(f1-f2), l2norm(f1+f2) from t2;
 print $rows , $data00 , $data01
 if $rows != 1 then
  return -1;
@ -117,7 +117,7 @@ if $data01 != 18.547236991 then
  return -1
 endi

-sql select sqr_sum(bit_and(f2, f1)), sqr_sum(bit_and(f1, f2)) from t2;
+sql select l2norm(bit_and(f2, f1)), l2norm(bit_and(f1, f2)) from t2;
 print $rows , $data00 , $data01
 if $rows != 1 then
  return -1
@ -129,7 +129,7 @@ if $data01 != 1.414213562 then
  return -1
 endi

-sql select sqr_sum(f2) from udf.t2 group by 1-bit_and(f1, f2) order by 1-bit_and(f1,f2);
+sql select l2norm(f2) from udf.t2 group by 1-bit_and(f1, f2) order by 1-bit_and(f1,f2);
 print $rows , $data00 , $data10 , $data20
 if $rows != 3 then
  return -1
@ -149,10 +149,10 @@ sql show functions;
 if $rows != 1 then
  return -1
 endi
-if $data00 != @sqr_sum@ then
+if $data00 != @l2norm@ then
  return -1
  endi
-sql drop function sqr_sum;
+sql drop function l2norm;
 sql show functions;
 if $rows != 0 then
  return -1
--- a/tests/script/tsim/valgrind/basic3.sim
+++ b/tests/script/tsim/valgrind/basic3.sim
@ -50,11 +50,11 @@ while $i < $tbNum
 endw 

 print =============== step3: tb
-sql select tbcol5 - tbcol3 from tb1

-print =============== step4: stb
-sql select tbcol5 - tbcol3 from stb
+sql select sum(tbcol) from stb partition by tbname interval(1s) slimit 1 soffset 1;
+sql select sum(tbcol) from stb partition by tbname interval(1s) slimit 2 soffset 4 limit 10 offset 1;

+_OVER:
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
 print =============== check
 $null=
--- a/tests/script/tsim/valgrind/basic4.sim
+++ b/tests/script/tsim/valgrind/basic4.sim
@ -22,41 +22,35 @@ if $data(1)[4] != ready then
  goto step1
 endi

-print =============== step2: create db
-sql create database d1 vgroups 2 buffer 3
-sql show databases
-sql use d1
-sql show vgroups
+$tbPrefix = tb
+$tbNum = 5
+$rowNum = 10

-print =============== step3: create show stable
-sql create table if not exists stb (ts timestamp, c1 int, c2 float, c3 double) tags (t1 int unsigned)
-sql show stables
-if $rows != 1 then 
-  return -1
-endi
+print =============== step2: prepare data
+sql create database db vgroups 2
+sql use db
+sql create table if not exists stb (ts timestamp, tbcol int, tbcol2 float, tbcol3 double, tbcol4 binary(30), tbcol5 binary(30)) tags (tgcol int unsigned)

-print =============== step4: create show table
-sql create table ct1 using stb tags(1000)
-sql create table ct2 using stb tags(2000)
-sql create table ct3 using stb tags(3000)
-sql show tables
-if $rows != 3 then 
-  return -1
-endi
+$i = 0
+while $i < $tbNum
+  $tb = $tbPrefix . $i
+  sql create table $tb using stb tags( $i )
+  $x = 0
+  while $x < $rowNum
+    $cc = $x * 60000
+    $ms = 1601481600000 + $cc
+    sql insert into $tb values ($ms , $x , $x , $x , "abcd1234=-+*" , "123456 0" ) 
+    $x = $x + 1
+  endw 

-print =============== step5: insert data (null / update)
-sql insert into ct1 values(now+0s, 10, 2.0, 3.0)
-sql insert into ct1 values(now+1s, 11, 2.1, NULL)(now+2s, -12, -2.2, -3.2)(now+3s, -13, -2.3, -3.3)
-sql insert into ct2 values(now+0s, 10, 2.0, 3.0)
-sql insert into ct2 values(now+1s, 11, 2.1, 3.1)(now+2s, -12, -2.2, -3.2)(now+3s, -13, -2.3, -3.3)
-sql insert into ct3 values('2021-01-01 00:00:00.000', NULL, NULL, 3.0)
-sql insert into ct3 values('2022-03-02 16:59:00.010', 3  , 4, 5), ('2022-03-02 16:59:00.010', 33 , 4, 5), ('2022-04-01 16:59:00.011', 4,  4, 5), ('2022-04-01 16:59:00.011', 6,  4, 5), ('2022-03-06 16:59:00.013', 8,  4, 5);
-sql insert into ct3 values('2022-03-02 16:59:00.010', 103, 1, 2), ('2022-03-02 16:59:00.010', 303, 3, 4), ('2022-04-01 16:59:00.011', 40, 5, 6), ('2022-04-01 16:59:00.011', 60, 4, 5), ('2022-03-06 16:59:00.013', 80, 4, 5);
-
-print =============== step6: query data=
-
-sql select * from stb where t1 between 1000 and 2500
+  $cc = $x * 60000
+  $ms = 1601481600000 + $cc
+  sql insert into $tb values ($ms , NULL , NULL , NULL , NULL , NULL ) 
+  $i = $i + 1
+endw 

+print =============== step3: tb
+sql select distinct(tbname), tgcol from stb;

 _OVER:
 system sh/exec.sh -n dnode1 -s stop -x SIGINT
--- a/tests/script/tsim/valgrind/checkError6.sim
+++ b/tests/script/tsim/valgrind/checkError6.sim
@ -111,6 +111,9 @@ sql select _wstart, count(*) from tb1 session(ts, 1m)
 sql select count(tbcol), avg(tbcol), max(tbcol), min(tbcol), count(tbcol) from stb where ts <= 1601481840000 and ts >= 1601481800000 partition by tgcol interval(1m) fill(value, 0)
 sql select tbcol5 - tbcol3 from stb

+sql select spread( tbcol2 )/44, spread(tbcol2), 0.204545455 * 44 from stb;
+sql select min(tbcol) * max(tbcol) /4, sum(tbcol2) * apercentile(tbcol2, 20), apercentile(tbcol2, 33) + 52/9 from stb;
+
 print =============== step5: explain
 sql explain analyze select ts from stb where -2;
 sql explain analyze select ts from tb1;
--- a/tests/script/tsim/valgrind/checkError8.sim
+++ b/tests/script/tsim/valgrind/checkError8.sim
@ -0,0 +1,152 @@
+system sh/stop_dnodes.sh
+system sh/deploy.sh -n dnode1 -i 1
+system sh/exec.sh -n dnode1 -s start -v
+sql connect
+
+sql drop database if exists cdb
+sql create database if not exists cdb 
+sql use cdb
+sql create table stb1 (ts timestamp, c1 int, c2 float, c3 bigint, c4 smallint, c5 tinyint, c6 double, c7 bool, c8 binary(10), c9 nchar(9)) TAGS(t1 int, t2 binary(10), t3 double)
+sql create table tb1 using stb1 tags(1,'1',1.0)
+sql create table tb2 using stb1 tags(2,'2',2.0)
+sql create table tb3 using stb1 tags(3,'3',3.0)
+sql create table tb4 using stb1 tags(4,'4',4.0)
+sql create table tb5 using stb1 tags(5,'5',5.0)
+sql create table tb6 using stb1 tags(6,'6',6.0)
+
+sql insert into tb1 values ('2021-05-05 18:19:00',1,1.0,1,1,1,1.0,true ,'1','1')
+sql insert into tb1 values ('2021-05-05 18:19:01',2,2.0,2,2,2,2.0,true ,'2','2')
+sql insert into tb1 values ('2021-05-05 18:19:02',3,3.0,3,3,3,3.0,false,'3','3')
+sql insert into tb1 values ('2021-05-05 18:19:03',4,4.0,4,4,4,4.0,false,'4','4')
+sql insert into tb1 values ('2021-05-05 18:19:04',11,11.0,11,11,11,11.0,true ,'11','11')
+sql insert into tb1 values ('2021-05-05 18:19:05',12,12.0,12,12,12,12.0,true ,'12','12')
+sql insert into tb1 values ('2021-05-05 18:19:06',13,13.0,13,13,13,13.0,false,'13','13')
+sql insert into tb1 values ('2021-05-05 18:19:07',14,14.0,14,14,14,14.0,false,'14','14')
+sql insert into tb2 values ('2021-05-05 18:19:08',21,21.0,21,21,21,21.0,true ,'21','21')
+sql insert into tb2 values ('2021-05-05 18:19:09',22,22.0,22,22,22,22.0,true ,'22','22')
+sql insert into tb2 values ('2021-05-05 18:19:10',23,23.0,23,23,23,23.0,false,'23','23')
+sql insert into tb2 values ('2021-05-05 18:19:11',24,24.0,24,24,24,24.0,false,'24','24')
+sql insert into tb3 values ('2021-05-05 18:19:12',31,31.0,31,31,31,31.0,true ,'31','31')
+sql insert into tb3 values ('2021-05-05 18:19:13',32,32.0,32,32,32,32.0,true ,'32','32')
+sql insert into tb3 values ('2021-05-05 18:19:14',33,33.0,33,33,33,33.0,false,'33','33')
+sql insert into tb3 values ('2021-05-05 18:19:15',34,34.0,34,34,34,34.0,false,'34','34')
+sql insert into tb4 values ('2021-05-05 18:19:16',41,41.0,41,41,41,41.0,true ,'41','41')
+sql insert into tb4 values ('2021-05-05 18:19:17',42,42.0,42,42,42,42.0,true ,'42','42')
+sql insert into tb4 values ('2021-05-05 18:19:18',43,43.0,43,43,43,43.0,false,'43','43')
+sql insert into tb4 values ('2021-05-05 18:19:19',44,44.0,44,44,44,44.0,false,'44','44')
+sql insert into tb5 values ('2021-05-05 18:19:20',51,51.0,51,51,51,51.0,true ,'51','51')
+sql insert into tb5 values ('2021-05-05 18:19:21',52,52.0,52,52,52,52.0,true ,'52','52')
+sql insert into tb5 values ('2021-05-05 18:19:22',53,53.0,53,53,53,53.0,false,'53','53')
+sql insert into tb5 values ('2021-05-05 18:19:23',54,54.0,54,54,54,54.0,false,'54','54')  
+sql insert into tb6 values ('2021-05-05 18:19:24',61,61.0,61,61,61,61.0,true ,'61','61')
+sql insert into tb6 values ('2021-05-05 18:19:25',62,62.0,62,62,62,62.0,true ,'62','62')
+sql insert into tb6 values ('2021-05-05 18:19:26',63,63.0,63,63,63,63.0,false,'63','63')
+sql insert into tb6 values ('2021-05-05 18:19:27',64,64.0,64,64,64,64.0,false,'64','64')
+sql insert into tb6 values ('2021-05-05 18:19:28',NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL)
+
+sql create table stb2 (ts timestamp, u1 int unsigned, u2 bigint unsigned, u3 smallint unsigned, u4 tinyint unsigned, ts2 timestamp) TAGS(t1 int unsigned, t2 bigint unsigned, t3 timestamp, t4 int)
+sql create table tb2_1 using stb2 tags(1,1,'2021-05-05 18:38:38',1)
+sql create table tb2_2 using stb2 tags(2,2,'2021-05-05 18:58:58',2)
+
+sql insert into tb2_1 values ('2021-05-05 18:19:00',1,2,3,4,'2021-05-05 18:28:01')
+sql insert into tb2_1 values ('2021-05-05 18:19:01',5,6,7,8,'2021-05-05 18:28:02')
+sql insert into tb2_1 values ('2021-05-05 18:19:02',2,2,3,4,'2021-05-05 18:28:03')
+sql insert into tb2_1 values ('2021-05-05 18:19:03',5,6,7,8,'2021-05-05 18:28:04')
+sql insert into tb2_1 values ('2021-05-05 18:19:04',3,2,3,4,'2021-05-05 18:28:05')
+sql insert into tb2_1 values ('2021-05-05 18:19:05',5,6,7,8,'2021-05-05 18:28:06')
+sql insert into tb2_1 values ('2021-05-05 18:19:06',4,2,3,4,'2021-05-05 18:28:07')
+sql insert into tb2_1 values ('2021-05-05 18:19:07',5,6,7,8,'2021-05-05 18:28:08')
+sql insert into tb2_1 values ('2021-05-05 18:19:08',5,2,3,4,'2021-05-05 18:28:09')
+sql insert into tb2_1 values ('2021-05-05 18:19:09',5,6,7,8,'2021-05-05 18:28:10')
+sql insert into tb2_1 values ('2021-05-05 18:19:10',6,2,3,4,'2021-05-05 18:28:11')
+sql insert into tb2_2 values ('2021-05-05 18:19:11',5,6,7,8,'2021-05-05 18:28:12')
+sql insert into tb2_2 values ('2021-05-05 18:19:12',7,2,3,4,'2021-05-05 18:28:13')
+sql insert into tb2_2 values ('2021-05-05 18:19:13',5,6,7,8,'2021-05-05 18:28:14')
+sql insert into tb2_2 values ('2021-05-05 18:19:14',8,2,3,4,'2021-05-05 18:28:15')
+sql insert into tb2_2 values ('2021-05-05 18:19:15',5,6,7,8,'2021-05-05 18:28:16')
+
+sql create table stb3 (ts timestamp, c1 int, c2 float, c3 bigint, c4 smallint, c5 tinyint, c6 double, c7 bool, c8 binary(10), c9 nchar(9)) TAGS(t1 int, t2 binary(10), t3 double)
+sql create table tb3_1 using stb3 tags(1,'1',1.0)
+sql create table tb3_2 using stb3 tags(2,'2',2.0)
+
+sql insert into tb3_1 values ('2021-01-05 18:19:00',1,1.0,1,1,1,1.0,true ,'1','1')
+sql insert into tb3_1 values ('2021-02-05 18:19:01',2,2.0,2,2,2,2.0,true ,'2','2')
+sql insert into tb3_1 values ('2021-03-05 18:19:02',3,3.0,3,3,3,3.0,false,'3','3')
+sql insert into tb3_1 values ('2021-04-05 18:19:03',4,4.0,4,4,4,4.0,false,'4','4')
+sql insert into tb3_1 values ('2021-05-05 18:19:28',5,NULL,5,NULL,5,NULL,true,NULL,'5')
+sql insert into tb3_1 values ('2021-06-05 18:19:28',NULL,6.0,NULL,6,NULL,6.0,NULL,'6',NULL)
+sql insert into tb3_1 values ('2021-07-05 18:19:28',NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL)
+sql insert into tb3_2 values ('2021-01-06 18:19:00',11,11.0,11,11,11,11.0,true ,'11','11')
+sql insert into tb3_2 values ('2021-02-06 18:19:01',12,12.0,12,12,12,12.0,true ,'12','12')
+sql insert into tb3_2 values ('2021-03-06 18:19:02',13,13.0,13,13,13,13.0,false,'13','13')
+sql insert into tb3_2 values ('2021-04-06 18:19:03',14,14.0,14,14,14,14.0,false,'14','14')
+sql insert into tb3_2 values ('2021-05-06 18:19:28',15,NULL,15,NULL,15,NULL,true,NULL,'15')
+sql insert into tb3_2 values ('2021-06-06 18:19:28',NULL,16.0,NULL,16,NULL,16.0,NULL,'16',NULL)
+sql insert into tb3_2 values ('2021-07-06 18:19:28',NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL)
+
+sql create table stb4 (ts timestamp, c1 int, c2 float, c3 bigint, c4 smallint, c5 tinyint, c6 double, c7 bool, c8 binary(10), c9 nchar(9),c10 binary(16300)) TAGS(t1 int, t2 binary(10), t3 double)
+sql create table tb4_0 using stb4 tags(0,'0',0.0)
+sql create table tb4_1 using stb4 tags(1,'1',1.0)
+sql create table tb4_2 using stb4 tags(2,'2',2.0)
+sql create table tb4_3 using stb4 tags(3,'3',3.0)
+sql create table tb4_4 using stb4 tags(4,'4',4.0)
+
+$i = 0
+$ts0 = 1625850000000
+$blockNum = 5
+$delta = 0
+$tbname0 = tb4_
+$a = 0
+$b = 200
+$c = 400
+while $i < $blockNum
+  $x = 0
+  $rowNum = 5
+  while $x < $rowNum
+    $ts = $ts0 + $x
+    $a = $a + 1
+    $b = $b + 1
+    $c = $c + 1
+    $d = $x / 10
+    $tin = $rowNum
+    $binary = 'binary . $c
+    $binary = $binary . '
+    $nchar = 'nchar . $c
+    $nchar = $nchar . '
+    $tbname = 'tb4_ . $i
+    $tbname = $tbname . '
+    sql insert into $tbname values ( $ts , $a , $b , $c , $d , $d , $c , true, $binary , $nchar , $binary )
+    $x = $x + 1
+  endw
+
+  $i = $i + 1
+  $ts0 = $ts0 + 259200000
+endw
+
+
+print ============== query
+sql select a.ts,a.c1,a.c8 from (select * from stb1 where c7=true) a, (select * from stb1 where c1 > 30) b where a.ts=b.ts and ((a.c1 > 50 and a.c1 < 60) or (b.c2 > 60));;
+sql select * from stb1 where (c6 > 3.0 or c6 < 60) and c6 > 50 and (c6 != 53 or c6 != 63);;
+sql select ts,c1 from stb1 where (c1 > 60 or c1 < 10 or (c1 > 20 and c1 < 30)) and ts > '2021-05-05 18:19:00.000' and ts < '2021-05-05 18:19:25.000' and c1 != 21 and c1 != 22 order by ts;
+sql select a.* from (select * from stb1 where c7=true) a, (select * from stb1 where c1 > 30) b where a.ts=b.ts and a.c1 > 50 order by ts;;
+sql select a.ts,b.ts,a.c1,b.u1,b.u2 from (select * from stb1) a, (select * from stb2) b where a.ts=b.ts and (a.c1 < 10 or a.c1 > 30) and (b.u1 < 5 or b.u1 > 5) order by ts;;
+sql select a.ts,b.ts,a.c1,b.u1,b.u2 from (select * from stb1) a, (select * from stb2) b where a.ts=b.ts and a.c1 < 30 and b.u1 > 1 and a.c1 > 10 and b.u1 < 8 and b.u1<>5 order by ts;;
+sql select tb1.ts,tb1.*,tb2_1.* from tb1, tb2_1 where tb1.ts=tb2_1.ts and tb1.ts >= '2021-05-05 18:19:03.000' and tb1.c7=false and tb2_1.u3>4 order by ts;;
+sql select stb1.ts,stb1.c1,stb1.t1,stb2.ts,stb2.u1,stb2.t4 from stb1, stb2 where stb1.ts=stb2.ts and stb1.t1 = stb2.t4 order by ts;;
+sql select count(*) from stb1 where tbname like 'tb%' or c1 > 0;;
+sql select * from stb1 where tbname like 'tb%' and (t1=1 or t2=2 or t3=3) and t1 > 2 order by ts;;
+
+_OVER:
+system sh/exec.sh -n dnode1 -s stop -x SIGINT
+print =============== check
+$null=
+
+system_content sh/checkValgrind.sh -n dnode1 
+print cmd return result ----> [ $system_content ]
+if $system_content > 2 then
+  return -1
+endi 
+
+if $system_content == $null then
+  return -1
+endi 
--- a/tools/taos-tools
+++ b/tools/taos-tools
@ -0,0 +1 @@
+Subproject commit 3c7dafeea3e558968165b73bee0f51024898e3da
				`@ -0,0 +1 @@`
				`Subproject commit 7ed7a97715388fa144718764d6bf20f9bfc29a12`
				`@ -0,0 +1 @@`
				`Subproject commit 3c7dafeea3e558968165b73bee0f51024898e3da`